+
+
An extensive Reinforcement Learning (RL) for Combinatorial Optimization (CO) benchmark. Our goal is to provide a unified framework for RL-based CO algorithms, and to facilitate reproducible research in this field, decoupling the science from the engineering.
@@ -27,7 +42,9 @@ RL4CO is built upon:
- [PyTorch Lightning](https://github.com/Lightning-AI/lightning): a lightweight PyTorch wrapper for high-performance AI research
- [Hydra](https://github.com/facebookresearch/hydra): a framework for elegantly configuring complex applications
-
+
+
+
We offer flexible and efficient implementations of the following policies:
- **Constructive**: learn to construct a solution from scratch
@@ -35,12 +52,16 @@ We offer flexible and efficient implementations of the following policies:
- _NonAutoregressive (NAR)_: learn to predict a heuristic, such as a heatmap, to then construct a solution
- **Improvement**: learn to improve an pre-existing solution
-
+
+
+
We provide several utilities and modularization. For example, we modularize reusable components such as _environment embeddings_ that can easily be swapped to [solve new problems](https://github.com/ai4co/rl4co/blob/main/examples/3-creating-new-env-model.ipynb).
-
+
+
+
## Getting started
@@ -178,13 +199,11 @@ We are also on [Slack](https://join.slack.com/t/rl4co/shared_invite/zt-1ytz2c1v4
If you find RL4CO valuable for your research or applied projects:
```bibtex
-@misc{berto2024rl4co,
+@article{berto2024rl4co,
title={{RL4CO: an Extensive Reinforcement Learning for Combinatorial Optimization Benchmark}},
- author={Federico Berto and Chuanbo Hua and Junyoung Park and Laurin Luttmann and Yining Ma and Fanchen Bu and Jiarui Wang and Haoran Ye and Minsu Kim and Sanghyeok Choi and Zepeda Gast and Andre Hottung and Jianan Zhou and Jieyi Bi and Yu Hu and Fei Liu and Hyeonah Kim and Jiwoo Son and Haeyeon Kim and Davide Angioni and Wouter Kool and Zhiguang Cao and Jie Zhang and Kijung Shin and Cathy Wu and Sungsoo Ahn and Guojie Song and Changhyun Kwon and Lin Xie and Jinkyoo Park},
+ author={Federico Berto and Chuanbo Hua and Junyoung Park and Laurin Luttmann and Yining Ma and Fanchen Bu and Jiarui Wang and Haoran Ye and Minsu Kim and Sanghyeok Choi and Nayeli Gast Zepeda and Andr\'e Hottung and Jianan Zhou and Jieyi Bi and Yu Hu and Fei Liu and Hyeonah Kim and Jiwoo Son and Haeyeon Kim and Davide Angioni and Wouter Kool and Zhiguang Cao and Jie Zhang and Kijung Shin and Cathy Wu and Sungsoo Ahn and Guojie Song and Changhyun Kwon and Lin Xie and Jinkyoo Park},
year={2024},
- eprint={2306.17100},
- archivePrefix={arXiv},
- primaryClass={cs.LG},
+ journal={arXiv preprint arXiv:2306.17100},
note={\url{https://github.com/ai4co/rl4co}}
}
```
@@ -202,6 +221,4 @@ We invite you to join our AI4CO community, an open research group in Artificial
-
-
-
+
\ No newline at end of file
diff --git a/configs/experiment/routing/mdpomo.yaml b/configs/experiment/routing/mdpomo.yaml
new file mode 100644
index 00000000..2de2a532
--- /dev/null
+++ b/configs/experiment/routing/mdpomo.yaml
@@ -0,0 +1,41 @@
+# @package _global_
+
+defaults:
+ - override /model: pomo.yaml
+ - override /env: cvrp.yaml
+ - override /callbacks: default.yaml
+ - override /trainer: default.yaml
+ - override /logger: wandb.yaml
+
+env:
+ generator_params:
+ num_loc: 50
+ loc_distribution: "mix_distribution"
+
+
+logger:
+ wandb:
+ project: "rl4co"
+ tags: ["mdpomo", "${env.name}"]
+ group: "${env.name}${env.generator_params.num_loc}"
+ name: "mdpomo-${env.name}${env.generator_params.num_loc}"
+
+
+model:
+ batch_size: 512
+ train_data_size: 10_000
+ val_data_size: 10_000
+ test_data_size: 10_000
+ optimizer_kwargs:
+ lr: 1e-4
+ weight_decay: 1e-6
+ lr_scheduler:
+ "MultiStepLR"
+ lr_scheduler_kwargs:
+ milestones: [9001]
+ gamma: 0.1
+
+trainer:
+ max_epochs: 10000
+
+seed: 1234
diff --git a/configs/experiment/routing/polynet.yaml b/configs/experiment/routing/polynet.yaml
new file mode 100644
index 00000000..709a18cf
--- /dev/null
+++ b/configs/experiment/routing/polynet.yaml
@@ -0,0 +1,44 @@
+# @package _global_
+
+defaults:
+ - override /model: polynet.yaml
+ - override /env: tsp.yaml
+ - override /callbacks: default.yaml
+ - override /trainer: default.yaml
+ - override /logger: wandb.yaml
+
+env:
+ generator_params:
+ num_loc: 50
+ check_solution: False # optimization
+
+logger:
+ wandb:
+ project: "rl4co"
+ tags: ["polynet", "${env.name}"]
+ group: "${env.name}${env.generator_params.num_loc}"
+ name: "polynet-${env.name}${env.generator_params.num_loc}-${model.k}"
+
+model:
+ k: 100
+ val_num_solutions: ${model.k}
+ batch_size: 512
+ val_batch_size: 1024
+ test_batch_size: 1024
+ train_data_size: 1_280_000
+ val_data_size: 10_000
+ test_data_size: 10_000
+ optimizer_kwargs:
+ lr: 1e-4
+ weight_decay: 1e-6
+ lr_scheduler:
+ "MultiStepLR"
+ lr_scheduler_kwargs:
+ milestones: [80, 95]
+ gamma: 0.1
+
+trainer:
+ max_epochs: 100
+
+seed: 1234
+
diff --git a/configs/experiment/scheduling/am-pomo.yaml b/configs/experiment/scheduling/am-pomo.yaml
index a3d2cde7..eb49e2da 100644
--- a/configs/experiment/scheduling/am-pomo.yaml
+++ b/configs/experiment/scheduling/am-pomo.yaml
@@ -14,6 +14,7 @@ model:
_target_: rl4co.models.L2DAttnPolicy
env_name: ${env.name}
scaling_factor: ${scaling_factor}
+ normalization: "batch"
batch_size: 64
num_starts: 10
num_augment: 0
diff --git a/configs/experiment/scheduling/am-ppo.yaml b/configs/experiment/scheduling/am-ppo.yaml
index c5d38eb1..f9e5d354 100644
--- a/configs/experiment/scheduling/am-ppo.yaml
+++ b/configs/experiment/scheduling/am-ppo.yaml
@@ -43,14 +43,8 @@ model:
batch_size: 128
val_batch_size: 512
test_batch_size: 64
- # Song et al use 1000 iterations over batches of 20 = 20_000
- # We train 10 epochs on a set of 2000 instance = 20_000
train_data_size: 2000
mini_batch_size: 512
- reward_scale: scale
- optimizer_kwargs:
- lr: 1e-4
env:
- stepwise_reward: True
- _torchrl_mode: True
\ No newline at end of file
+ stepwise_reward: True
\ No newline at end of file
diff --git a/configs/experiment/scheduling/base.yaml b/configs/experiment/scheduling/base.yaml
index e84f95fd..c15a6c45 100644
--- a/configs/experiment/scheduling/base.yaml
+++ b/configs/experiment/scheduling/base.yaml
@@ -22,17 +22,19 @@ trainer:
seed: 12345678
-scaling_factor: 20
+scaling_factor: ${env.generator_params.max_processing_time}
model:
_target_: ???
batch_size: ???
train_data_size: 2_000
val_data_size: 1_000
- test_data_size: 1_000
+ test_data_size: 100
optimizer_kwargs:
- lr: 1e-4
+ lr: 2e-4
weight_decay: 1e-6
lr_scheduler: "ExponentialLR"
lr_scheduler_kwargs:
gamma: 0.95
+ reward_scale: scale
+ max_grad_norm: 1
diff --git a/configs/experiment/scheduling/ffsp-matnet.yaml b/configs/experiment/scheduling/ffsp-matnet.yaml
new file mode 100644
index 00000000..7a1f4148
--- /dev/null
+++ b/configs/experiment/scheduling/ffsp-matnet.yaml
@@ -0,0 +1,46 @@
+# @package _global_
+
+defaults:
+ - override /model: matnet.yaml
+ - override /callbacks: default.yaml
+ - override /trainer: default.yaml
+ - override /logger: wandb.yaml
+ - override /env: ffsp.yaml
+
+logger:
+ wandb:
+ project: "rl4co"
+ log_model: "all"
+ group: "${env.name}-${env.generator_params.num_job}-${env.generator_params.num_machine}"
+ tags: ["matnet", "${env.name}"]
+ name: "matnet-${env.name}-${env.generator_params.num_job}j-${env.generator_params.num_machine}m"
+
+env:
+ generator_params:
+ num_stage: 3
+ num_machine: 4
+ num_job: 20
+ flatten_stages: False
+
+trainer:
+ max_epochs: 50
+ # NOTE for some reason l2d is extremely sensitive to precision
+ # ONLY USE 32-true for l2d!
+ precision: 32-true
+ gradient_clip_val: 10 # orig paper does not use grad clipping
+
+seed: 12345678
+
+model:
+ batch_size: 50
+ train_data_size: 10_000
+ val_data_size: 1_000
+ test_data_size: 1_000
+ optimizer_kwargs:
+ lr: 1e-4
+ weight_decay: 1e-6
+ lr_scheduler:
+ "MultiStepLR"
+ lr_scheduler_kwargs:
+ milestones: [35, 45]
+ gamma: 0.1
diff --git a/configs/experiment/scheduling/gnn-ppo.yaml b/configs/experiment/scheduling/gnn-ppo.yaml
index d9c04856..d2139eea 100644
--- a/configs/experiment/scheduling/gnn-ppo.yaml
+++ b/configs/experiment/scheduling/gnn-ppo.yaml
@@ -12,24 +12,22 @@ logger:
model:
_target_: rl4co.models.L2DPPOModel
policy_kwargs:
- embed_dim: 128
+ embed_dim: 256
num_encoder_layers: 3
scaling_factor: ${scaling_factor}
- max_grad_norm: 1
- ppo_epochs: 3
+ ppo_epochs: 2
het_emb: False
+ normalization: instance
+ test_decode_type: greedy
batch_size: 128
val_batch_size: 512
test_batch_size: 64
mini_batch_size: 512
- reward_scale: scale
- optimizer_kwargs:
- lr: 1e-4
+
trainer:
max_epochs: 10
env:
- stepwise_reward: True
- _torchrl_mode: True
\ No newline at end of file
+ stepwise_reward: True
\ No newline at end of file
diff --git a/configs/experiment/scheduling/hgnn-pomo.yaml b/configs/experiment/scheduling/hgnn-pomo.yaml
index eb688c03..a964143f 100644
--- a/configs/experiment/scheduling/hgnn-pomo.yaml
+++ b/configs/experiment/scheduling/hgnn-pomo.yaml
@@ -18,6 +18,7 @@ model:
stepwise_encoding: False
scaling_factor: ${scaling_factor}
het_emb: True
+ normalization: instance
num_starts: 10
batch_size: 64
num_augment: 0
diff --git a/configs/experiment/scheduling/hgnn-ppo.yaml b/configs/experiment/scheduling/hgnn-ppo.yaml
index 8e3a62d8..7d46f7d7 100644
--- a/configs/experiment/scheduling/hgnn-ppo.yaml
+++ b/configs/experiment/scheduling/hgnn-ppo.yaml
@@ -12,24 +12,16 @@ logger:
model:
_target_: rl4co.models.L2DPPOModel
policy_kwargs:
- embed_dim: 128
+ embed_dim: 256
num_encoder_layers: 3
scaling_factor: ${scaling_factor}
- max_grad_norm: 1
- ppo_epochs: 3
+ ppo_epochs: 2
het_emb: True
+ normalization: instance
batch_size: 128
val_batch_size: 512
test_batch_size: 64
mini_batch_size: 512
- reward_scale: scale
- optimizer_kwargs:
- lr: 1e-4
-
-trainer:
- max_epochs: 10
-
env:
- stepwise_reward: True
- _torchrl_mode: True
\ No newline at end of file
+ stepwise_reward: True
\ No newline at end of file
diff --git a/configs/experiment/scheduling/matnet-ppo.yaml b/configs/experiment/scheduling/matnet-ppo.yaml
index f0e30e3b..c88d2c64 100644
--- a/configs/experiment/scheduling/matnet-ppo.yaml
+++ b/configs/experiment/scheduling/matnet-ppo.yaml
@@ -36,13 +36,7 @@ model:
batch_size: 128
val_batch_size: 512
test_batch_size: 64
- # Song et al use 1000 iterations over batches of 20 = 20_000
- # We train 10 epochs on a set of 2000 instance = 20_000
mini_batch_size: 512
- reward_scale: scale
- optimizer_kwargs:
- lr: 1e-4
env:
- stepwise_reward: True
- _torchrl_mode: True
\ No newline at end of file
+ stepwise_reward: True
\ No newline at end of file
diff --git a/configs/model/polynet.yaml b/configs/model/polynet.yaml
new file mode 100644
index 00000000..4070d120
--- /dev/null
+++ b/configs/model/polynet.yaml
@@ -0,0 +1,11 @@
+_target_: rl4co.models.PolyNet
+
+# During val/test only we apply dihedral transformations
+num_augment: 8
+
+metrics:
+ train: ["loss", "reward"]
+ val: ["reward", "max_reward", "max_aug_reward"]
+ test: ${model.metrics.val}
+ log_on_step: True
+
diff --git a/docs/Makefile b/docs/Makefile
deleted file mode 100644
index d4bb2cbb..00000000
--- a/docs/Makefile
+++ /dev/null
@@ -1,20 +0,0 @@
-# Minimal makefile for Sphinx documentation
-#
-
-# You can set these variables from the command line, and also
-# from the environment for the first two.
-SPHINXOPTS ?=
-SPHINXBUILD ?= sphinx-build
-SOURCEDIR = .
-BUILDDIR = _build
-
-# Put it first so that "make" without argument is like "make help".
-help:
- @$(SPHINXBUILD) -M help "$(SOURCEDIR)" "$(BUILDDIR)" $(SPHINXOPTS) $(O)
-
-.PHONY: help Makefile
-
-# Catch-all target: route all unknown targets to Sphinx using the new
-# "make mode" option. $(O) is meant as a shortcut for $(SPHINXOPTS).
-%: Makefile
- @$(SPHINXBUILD) -M $@ "$(SOURCEDIR)" "$(BUILDDIR)" $(SPHINXOPTS) $(O)
diff --git a/docs/README.md b/docs/README.md
index b89e38ec..169ab0e3 100644
--- a/docs/README.md
+++ b/docs/README.md
@@ -1,114 +1,26 @@
-# π RL4CO Docs
+# RL4CO Documentation
-We are using [Sphinx](https://www.sphinx-doc.org/en/master/) with [Napoleon extension](https://www.sphinx-doc.org/en/master/usage/extensions/napoleon.html) to build the documentation.
-Moreover, we set [Google style](https://google.github.io/styleguide/pyguide.html) to follow with type convention.
+We use [MkDocs](https://www.mkdocs.org/) to generate the documentation with the [MkDocs Material theme](https://squidfunk.github.io/mkdocs-material/).
-- [Napoleon formatting with Google style](https://sphinxcontrib-napoleon.readthedocs.io/en/latest/example_google.html)
-- [ReStructured Text (reST)](https://docs.pylonsproject.org/projects/docs-style-guide/)
-- [Paragraph-level markup](https://www.sphinx-doc.org/en/master/usage/restructuredtext/basics.html#paragraphs)
+## Development
-See following short example of a sample function taking one position string and optional
+From the root directory:
-```python
-from typing import Optional
+1. Install RL4CO locally
-
-def my_func(param_a: int, param_b: Optional[float] = None) -> str:
- """Sample function.
-
- Args:
- param_a: first parameter
- param_b: second parameter
-
- Return:
- sum of both numbers
-
- Example::
-
- >>> my_func(1, 2)
- 3
-
- Note:
- If you want to add something.
- """
- p = param_b if param_b else 0
- return str(param_a + p)
+```bash
+pip install -e ".[dev,graph,routing,docs]"
```
-## ποΈ File structures
-
-```
-.
-βββ _build/ - output website, only for local building
-βββ _content/ - content for docs pages in Markdown format
-βββ _theme/
-β βββ rl4co/ - website theme files
-βββ conf.py - main config file for building
-βββ index.md - content for index page
-βββ make.bat - building script for Windows
-βββ Makefile - building script for Unix
-βββ README.md
-βββ requirements.txt - requirement python packages for Read the Docs building
-```
-
-## βοΈ Build docs locally
-
-**Step 1**. Install requirement packages (from root folder): `pip install -r docs/requirement.txt`;
-
-**Step 2**. Run the building script:
+note that `docs` is the extra requirement for the documentation.
-- **Windows**: run `make.bat`;
-- **Linux/macOS**: run `make html`;
-The generated docs will be under the `_build` folder. You can open `docs/build/html/index.html` in your browser to check the docs.
+2. To build the documentation, run:
-We need to have LaTeX installed for rendering math equations. You can for example install TeXLive with the necessary extras by doing one of the following:
-
-- **Windows/macOS**: check the [Tex Live install guide](https://www.tug.org/texlive/windows.html) for Windows/macOS;
-- **Ubuntu (Linux)**: run `sudo apt-get update && sudo apt-get install -y texlive-latex-extra dvipng texlive-pictures`;
-- Use the [RTD docker image](https://hub.docker.com/r/readthedocs/build);
-
-## βοΈ Build in Read the Docs
-
-In the root of this repository, there is `.readthedocs.yaml` which will be loaded by the Read the Docs to build the docs. Please refer to the [configuration file v2 guide from Read the Docs](https://docs.readthedocs.io/en/stable/config-file/v2.html) for details information of variables.
-
-## π‘ Notes for contents
-
-Markdown and RST support
-RST is originally supported by the Sphinx. With the extension `myst_parser` it can support Markdown contents. Follow [this guide](https://www.sphinx-doc.org/en/master/usage/markdown.html) to learn more.
-
-In the meantime, we can still use RST within Markdown files by
-````
-```{eval-rst}
-RST CONTENTS
-```
-````
-
-
-Jupyter notebook support
-With the extension `nbsphinx`, Sphinx can support Jupyter notebook. Follow [this guide](https://docs.readthedocs.io/en/stable/guides/jupyter.html) to learn more.
-
-Indexing a Jupyter notebook is the same with a Markdown file in RST:
-```
-.. toctree::
- :maxdepth: 2
- :caption: Getting started:
-
- _content/start/installation
- _content/start/quickstart_notebook
-```
-
-
-API docs auto generator
-With Sphinx's `automodule` we can easily get the API docs:
-```
-.. automodule:: rl4co.data.generate_data
- :members:
- :undoc-members:
+```bash
+mkdocs serve
```
-When deploy in Read the Docs, make sure putting the package to `requirement.txt` mentioned before.
-
-## π References
+### Hooks
-We base the above guide on the official [PyTorch Lightning Docs](https://github.com/Lightning-AI/lightning/tree/master/docs).
\ No newline at end of file
+We are using the [hooks.py](hooks.py) for additional modifications. MkDocs for instance cannot detect files that are not in the same directory as an `__init__.py` (as described [here](https://stackoverflow.com/questions/75232397/mkdocs-unable-to-find-modules)) so we are automatically creating and deleting such files with our script
diff --git a/docs/_content/api/algos/a2c.md b/docs/_content/api/algos/a2c.md
deleted file mode 100644
index 731f48bc..00000000
--- a/docs/_content/api/algos/a2c.md
+++ /dev/null
@@ -1,9 +0,0 @@
-# A2C
-
-## A2C (Advantage Actor Critic)
-
-```{eval-rst}
-.. automodule:: rl4co.models.rl.a2c.a2c
- :members:
- :undoc-members:
-```
diff --git a/docs/_content/api/algos/base.md b/docs/_content/api/algos/base.md
deleted file mode 100644
index 243c9dab..00000000
--- a/docs/_content/api/algos/base.md
+++ /dev/null
@@ -1,9 +0,0 @@
-# Base Classes
-
-## RL4COLitModule
-
-```{eval-rst}
-.. automodule:: rl4co.models.rl.common.base
- :members:
- :undoc-members:
-```
\ No newline at end of file
diff --git a/docs/_content/api/algos/ppo.md b/docs/_content/api/algos/ppo.md
deleted file mode 100644
index d99fd1d3..00000000
--- a/docs/_content/api/algos/ppo.md
+++ /dev/null
@@ -1,11 +0,0 @@
-# PPO
-
-
-## PPO (Proximal Policy Optimization)
-
-
-```{eval-rst}
-.. automodule:: rl4co.models.rl.ppo.ppo
- :members:
- :undoc-members:
-```
\ No newline at end of file
diff --git a/docs/_content/api/algos/reinforce.md b/docs/_content/api/algos/reinforce.md
deleted file mode 100644
index ea8070b1..00000000
--- a/docs/_content/api/algos/reinforce.md
+++ /dev/null
@@ -1,17 +0,0 @@
-# Reinforce
-
-## REINFORCE
-
-```{eval-rst}
-.. automodule:: rl4co.models.rl.reinforce.reinforce
- :members:
- :undoc-members:
-```
-
-## Baselines
-
-```{eval-rst}
-.. automodule:: rl4co.models.rl.reinforce.baselines
- :members:
- :undoc-members:
-```
\ No newline at end of file
diff --git a/docs/_content/api/data.md b/docs/_content/api/data.md
deleted file mode 100644
index 87690422..00000000
--- a/docs/_content/api/data.md
+++ /dev/null
@@ -1,33 +0,0 @@
-# Data
-
-## Datasets
-
-```{eval-rst}
-.. automodule:: rl4co.data.dataset
- :members:
- :undoc-members:
-```
-
-## Data Generation
-
-```{eval-rst}
-.. automodule:: rl4co.data.generate_data
- :members:
- :undoc-members:
-```
-
-## Transforms
-
-```{eval-rst}
-.. automodule:: rl4co.data.transforms
- :members:
- :undoc-members:
-```
-
-## Utils
-
-```{eval-rst}
-.. automodule:: rl4co.data.utils
- :members:
- :undoc-members:
-```
\ No newline at end of file
diff --git a/docs/_content/api/decoding.md b/docs/_content/api/decoding.md
deleted file mode 100644
index 1ba9470b..00000000
--- a/docs/_content/api/decoding.md
+++ /dev/null
@@ -1,7 +0,0 @@
-# Decoding Strategies
-
-```{eval-rst}
-.. automodule:: rl4co.utils.decoding
- :members:
- :undoc-members:
-```
diff --git a/docs/_content/api/envs/eda.md b/docs/_content/api/envs/eda.md
deleted file mode 100644
index d89568b1..00000000
--- a/docs/_content/api/envs/eda.md
+++ /dev/null
@@ -1,30 +0,0 @@
-# EDA Problems
-Environment for Electronic Design Automation (EDA) problems
-
-## Decap Placement Problem (DPP)
-
-```{eval-rst}
-.. automodule:: rl4co.envs.eda.dpp.env
- :members:
- :undoc-members:
-```
-
-```{eval-rst}
-.. automodule:: rl4co.envs.eda.dpp.generator
- :members:
- :undoc-members:
-```
-
-## Multi-port Decap Placement Problem (mDPP)
-
-```{eval-rst}
-.. automodule:: rl4co.envs.eda.mdpp.env
- :members:
- :undoc-members:
-```
-
-```{eval-rst}
-.. automodule:: rl4co.envs.eda.mdpp.generator
- :members:
- :undoc-members:
-```
diff --git a/docs/_content/api/envs/routing.md b/docs/_content/api/envs/routing.md
deleted file mode 100644
index 2ee370cb..00000000
--- a/docs/_content/api/envs/routing.md
+++ /dev/null
@@ -1,139 +0,0 @@
-# Routing Problems
-
-See also the [Multi-Task VRP](#Multi-Task-Vehicle-Routing-Problem-(MTVRP)) at the bottom of this page, that includes 16 variants!
-
-
-
-## Asymmetric Traveling Salesman Problem (ATSP)
-
-```{eval-rst}
-.. automodule:: rl4co.envs.routing.atsp.env
- :members:
- :undoc-members:
-```
-
-```{eval-rst}
-.. automodule:: rl4co.envs.routing.atsp.generator
- :members:
- :undoc-members:
-```
-
-## Capacitated Vehicle Routing Problem (CVRP)
-
-```{eval-rst}
-.. automodule:: rl4co.envs.routing.cvrp.env
- :members:
- :undoc-members:
-```
-
-```{eval-rst}
-.. automodule:: rl4co.envs.routing.cvrp.generator
- :members:
- :undoc-members:
-```
-
-## Multiple Traveling Salesman Problem (mTSP)
-
-```{eval-rst}
-.. automodule:: rl4co.envs.routing.mtsp.env
- :members:
- :undoc-members:
-```
-
-```{eval-rst}
-.. automodule:: rl4co.envs.routing.mtsp.generator
- :members:
- :undoc-members:
-```
-
-
-## Orienteering Problem (OP)
-
-```{eval-rst}
-.. automodule:: rl4co.envs.routing.op.env
- :members:
- :undoc-members:
-```
-
-```{eval-rst}
-.. automodule:: rl4co.envs.routing.op.generator
- :members:
- :undoc-members:
-```
-
-## Pickup and Delivery Problem (PDP)
-
-```{eval-rst}
-.. automodule:: rl4co.envs.routing.pdp.env
- :members:
- :undoc-members:
-```
-
-```{eval-rst}
-.. automodule:: rl4co.envs.routing.pdp.generator
- :members:
- :undoc-members:
-```
-
-## Prize Collecting Traveling Salesman Problem (PCTSP)
-
-```{eval-rst}
-.. automodule:: rl4co.envs.routing.pctsp.env
- :members:
- :undoc-members:
-```
-
-```{eval-rst}
-.. automodule:: rl4co.envs.routing.pctsp.generator
- :members:
- :undoc-members:
-```
-
-## Split Delivery Vehicle Routing Problem (SDVRP)
-
-```{eval-rst}
-.. automodule:: rl4co.envs.routing.sdvrp.env
- :members:
- :undoc-members:
-```
-```
-
-## Stochastic Prize Collecting Traveling Salesman Problem (SPCTSP)
-
-```{eval-rst}
-.. automodule:: rl4co.envs.routing.spctsp.env
- :members:
- :undoc-members:
-```
-```
-
-## Traveling Salesman Problem (TSP)
-
-```{eval-rst}
-.. automodule:: rl4co.envs.routing.tsp.env
- :members:
- :undoc-members:
-```
-
-```{eval-rst}
-.. automodule:: rl4co.envs.routing.tsp.generator
- :members:
- :undoc-members:
-```
-
----
-
-
-## Multi-Task Vehicle Routing Problem (MTVRP)
-
-
-
-```{eval-rst}
-.. automodule:: rl4co.envs.routing.mtvrp.env
- :members:
- :undoc-members:
-
-.. automodule:: rl4co.envs.routing.mtvrp.generator
- :members:
- :undoc-members:
-```
\ No newline at end of file
diff --git a/docs/_content/api/envs/scheduling.md b/docs/_content/api/envs/scheduling.md
deleted file mode 100644
index 46edb20d..00000000
--- a/docs/_content/api/envs/scheduling.md
+++ /dev/null
@@ -1,29 +0,0 @@
-# Scheduling Problems
-
-## Flexible Flow Shop Problem (FFSP)
-
-```{eval-rst}
-.. automodule:: rl4co.envs.scheduling.ffsp.env
- :members:
- :undoc-members:
-```
-
-```{eval-rst}
-.. automodule:: rl4co.envs.scheduling.ffsp.generator
- :members:
- :undoc-members:
-```
-
-## Single Machine Total Weighted Tardiness Problem (SMTWTP)
-
-```{eval-rst}
-.. automodule:: rl4co.envs.scheduling.smtwtp.env
- :members:
- :undoc-members:
-```
-
-```{eval-rst}
-.. automodule:: rl4co.envs.scheduling.smtwtp.generator
- :members:
- :undoc-members:
-```
\ No newline at end of file
diff --git a/docs/_content/api/models/common/__init__.md b/docs/_content/api/models/common/__init__.md
deleted file mode 100644
index bba01830..00000000
--- a/docs/_content/api/models/common/__init__.md
+++ /dev/null
@@ -1,54 +0,0 @@
-# NCO Methods Overview
-
-
-We categorize NCO approaches (which are in fact not necessarily trained with RL!) into the following: 1) constructive, 2) improvement, 3) transductive.
-
-
-
-```{eval-rst}
-.. tip::
- Note that in RL4CO we distinguish the RL algorithms and the actors via the following naming:
-
- * **Model:** Refers to the reinforcement learning algorithm encapsulated within a `LightningModule`. This module is responsible for training the policy.
- * **Policy:** Implemented as a `nn.Module`, this neural network (often referred to as the *actor*) takes an instance and outputs a sequence of actions, :math:`\pi = \pi_0, \pi_1, \dots, \pi_N`, which constitutes the solution.
-
- Here, :math:`\pi_i` represents the action taken at step :math:`i`, forming a sequence that leads to the optimal or near-optimal solution for the given instance.
-```
-
-
-The following table contains the categorization that we follow in RL4CO:
-
-
-```{eval-rst}
-.. list-table:: Overview of RL Models and Policies
- :widths: 5 5 5 5 25
- :header-rows: 1
- :stub-columns: 1
-
- * - Category
- - Model or Policy?
- - Input
- - Output
- - Description
- * - `Constructive `_
- - Policy
- - Instance
- - Solution
- - Policies trained to generate solutions from scratch. Can be categorized into AutoRegressive (AR) and Non-Autoregressive (NAR).
- * - `Improvement `_
- - Policy
- - Instance, Current Solution
- - Improved Solution
- - Policies trained to improve existing solutions iteratively, akin to local search algorithms. They focus on refining *existing* solutions rather than generating them from scratch.
- * - `Transductive `_
- - Model
- - Instance, (Policy)
- - Solution, (Updated Policy)
- - Updates policy parameters during online testing to improve solutions of a specific instance.
-```
-
-
-
-
-
-
diff --git a/docs/_content/api/models/common/constructive.md b/docs/_content/api/models/common/constructive.md
deleted file mode 100644
index 4b3d0e82..00000000
--- a/docs/_content/api/models/common/constructive.md
+++ /dev/null
@@ -1,71 +0,0 @@
-## Constructive Policies
-
-Constructive NCO policies pre-train a policy to amortize the inference. "Constructive" means that a solution is created from scratch by the model. We can also categorize constructive NCO in two sub-categories depending on the role of encoder and decoder:
-
-#### Autoregressive (AR)
-Autoregressive approaches **use a learned decoder** that outputs log probabilities for the current solution. These approaches generate a solution step by step, similar to e.g. LLMs. They have an encoder-decoder structure. Some models may not have an encoder at all and just re-encode at each step.
-
-#### NonAutoregressive (NAR)
-The difference between AR and NAR approaches is that NAR **only an encoder is learnable** (they just encode in one shot) and generate for example a heatmap, which can then be decoded simply by using it as a probability distribution or by using some search method on top.
-
-Here is a general structure of a general constructive policy with an encoder-decoder structure:
-
-
-
-
-where _embeddings_ transfer information from feature space to embedding space.
-
----
-
-
-
-### Constructive Policy Base Classes
-
-```{eval-rst}
-.. automodule:: rl4co.models.common.constructive.base
- :members:
- :undoc-members:
-```
-
-
-
-### Autoregressive Policies Base Classes
-
-```{eval-rst}
-.. automodule:: rl4co.models.common.constructive.autoregressive.encoder
- :members:
- :undoc-members:
-```
-
-```{eval-rst}
-.. automodule:: rl4co.models.common.constructive.autoregressive.decoder
- :members:
- :undoc-members:
-```
-
-```{eval-rst}
-.. automodule:: rl4co.models.common.constructive.autoregressive.policy
- :members:
- :undoc-members:
-```
-
-### Nonautoregressive Policies Base Classes
-
-
-```{eval-rst}
-.. automodule:: rl4co.models.common.constructive.nonautoregressive.encoder
- :members:
- :undoc-members:
-```
-
-```{eval-rst}
-.. automodule:: rl4co.models.common.constructive.nonautoregressive.decoder
- :members:
- :undoc-members:
-```
-
-```{eval-rst}
-.. automodule:: rl4co.models.common.constructive.nonautoregressive.policy
- :members:
- :undoc-members:
-```
diff --git a/docs/_content/api/models/common/improvement.md b/docs/_content/api/models/common/improvement.md
deleted file mode 100644
index 80543b2a..00000000
--- a/docs/_content/api/models/common/improvement.md
+++ /dev/null
@@ -1,3 +0,0 @@
-## Improvement Policies
-
-These methods differ w.r.t. constructive NCO since they can obtain better solutions similarly to how local search algorithms work - they can improve the solutions over time. This is different from decoding strategies or similar in constructive methods since these policies are trained for performing improvement operations.
diff --git a/docs/_content/api/models/common/transductive.md b/docs/_content/api/models/common/transductive.md
deleted file mode 100644
index 6c14c036..00000000
--- a/docs/_content/api/models/common/transductive.md
+++ /dev/null
@@ -1,19 +0,0 @@
-# Transductive Models
-
-
-Transductive models are learning algorithms that optimize on a specific instance. They improve solutions by updating policy parameters $\theta$_, which means that we are running optimization (backprop) **at test time**. Transductive learning can be performed with different policies: for example EAS updates (a part of) AR policies parameters to obtain better solutions, but I guess there are ways (or papers out there I don't know of) that optimize at test time.
-
-
-```{eval-rst}
-.. tip::
- You may refer to the definition of `inductive vs transductive RL `_. In inductive RL, we train to generalize to new instances. In transductive RL we train (or finetune) to solve only specific ones.
-```
-
-
-## Base Transductive Model
-
-```{eval-rst}
-.. automodule:: rl4co.models.common.transductive.base
- :members:
- :undoc-members:
-```
\ No newline at end of file
diff --git a/docs/_content/api/models/nn.md b/docs/_content/api/models/nn.md
deleted file mode 100644
index ba5fa86b..00000000
--- a/docs/_content/api/models/nn.md
+++ /dev/null
@@ -1,60 +0,0 @@
-# Neural Network Modules
-
-## Critic Network
-
-```{eval-rst}
-.. automodule:: rl4co.models.rl.common.critic
- :members:
- :undoc-members:
-```
-
-## Graph Neural Networks
-
-### Graph Attention Encoder
-
-```{eval-rst}
-.. automodule:: rl4co.models.nn.graph.attnnet
- :members:
- :undoc-members:
-```
-
-### Graph Convolutional Encoder
-
-```{eval-rst}
-.. automodule:: rl4co.models.nn.graph.gcn
- :members:
- :undoc-members:
-```
-
-### Message Passing Encoder
-
-```{eval-rst}
-.. automodule:: rl4co.models.nn.graph.mpnn
- :members:
- :undoc-members:
-```
-
-## rl4co.models.nn.attention
-
-```{eval-rst}
-.. automodule:: rl4co.models.nn.attention
- :members:
- :undoc-members:
-```
-
-
-## rl4co.models.nn.mlp
-
-```{eval-rst}
-.. automodule:: rl4co.models.nn.mlp
- :members:
- :undoc-members:
-```
-
-## rl4co.models.nn.ops
-
-```{eval-rst}
-.. automodule:: rl4co.models.nn.ops
- :members:
- :undoc-members:
-```
\ No newline at end of file
diff --git a/docs/_content/api/models/zoo/__init__.md b/docs/_content/api/models/zoo/__init__.md
deleted file mode 100644
index dee807f8..00000000
--- a/docs/_content/api/models/zoo/__init__.md
+++ /dev/null
@@ -1,3 +0,0 @@
-# Model Zoo
-
-Models from the literature and contributions are contained in the Model Zoo.
\ No newline at end of file
diff --git a/docs/_content/api/models/zoo/constructive.md b/docs/_content/api/models/zoo/constructive.md
deleted file mode 100644
index 4a1b1993..00000000
--- a/docs/_content/api/models/zoo/constructive.md
+++ /dev/null
@@ -1,215 +0,0 @@
-# Constructive Methods
-
-These can be classified into Auto-Regressive (AR) and Non-Auto-Regressive (NAR) policies. These constructive methods also include ad-hoc RL algorithms for constructive policies.
-
-# AutoRegressive
-
-## Attention Model (AM)
-
-```{eval-rst}
-.. automodule:: rl4co.models.zoo.am.model
- :members:
- :undoc-members:
-```
-
-```{eval-rst}
-.. automodule:: rl4co.models.zoo.am.policy
- :members:
- :undoc-members:
-```
-
-## Attention Model - PPO (AM-PPO)
-
-```{eval-rst}
-.. automodule:: rl4co.models.zoo.amppo.model
- :members:
- :undoc-members:
-```
-
-
-
-## Heterogeneous Attention Model (HAM)
-
-```{eval-rst}
-.. automodule:: rl4co.models.zoo.ham.model
- :members:
- :undoc-members:
-```
-
-```{eval-rst}
-.. automodule:: rl4co.models.zoo.ham.policy
- :members:
- :undoc-members:
-```
-
-```{eval-rst}
-.. automodule:: rl4co.models.zoo.ham.encoder
- :members:
- :undoc-members:
-```
-
-```{eval-rst}
-.. automodule:: rl4co.models.zoo.ham.attention
- :members:
- :undoc-members:
-```
-
-
-## Matrix Encoding Network (MatNet)
-
-```{eval-rst}
-.. automodule:: rl4co.models.zoo.matnet.model
- :members:
- :undoc-members:
-```
-
-```{eval-rst}
-.. automodule:: rl4co.models.zoo.matnet.policy
- :members:
- :undoc-members:
-```
-
-```{eval-rst}
-.. automodule:: rl4co.models.zoo.matnet.encoder
- :members:
- :undoc-members:
-```
-
-```{eval-rst}
-.. automodule:: rl4co.models.zoo.matnet.decoder
- :members:
- :undoc-members:
-```
-
-
-## Multi-Decoder Attention Model (MDAM)
-
-```{eval-rst}
-.. automodule:: rl4co.models.zoo.mdam.model
- :members:
- :undoc-members:
-```
-
-```{eval-rst}
-.. automodule:: rl4co.models.zoo.mdam.policy
- :members:
- :undoc-members:
-```
-
-```{eval-rst}
-.. automodule:: rl4co.models.zoo.mdam.encoder
- :members:
- :undoc-members:
-```
-
-```{eval-rst}
-.. automodule:: rl4co.models.zoo.mdam.decoder
- :members:
- :undoc-members:
-```
-
-## POMO
-
-```{eval-rst}
-.. automodule:: rl4co.models.zoo.pomo.model
- :members:
- :undoc-members:
-```
-
-## Pointer Network (PtrNet)
-
-```{eval-rst}
-.. automodule:: rl4co.models.zoo.ptrnet.model
- :members:
- :undoc-members:
-```
-
-```{eval-rst}
-.. automodule:: rl4co.models.zoo.ptrnet.policy
- :members:
- :undoc-members:
-```
-
-```{eval-rst}
-.. automodule:: rl4co.models.zoo.ptrnet.encoder
- :members:
- :undoc-members:
-```
-
-```{eval-rst}
-.. automodule:: rl4co.models.zoo.ptrnet.decoder
- :members:
- :undoc-members:
-```
-
-```{eval-rst}
-.. automodule:: rl4co.models.zoo.ptrnet.critic
- :members:
- :undoc-members:
-```
-
-## SymNCO
-
-```{eval-rst}
-.. automodule:: rl4co.models.zoo.symnco.model
- :members:
- :undoc-members:
-```
-
-```{eval-rst}
-.. automodule:: rl4co.models.zoo.symnco.policy
- :members:
- :undoc-members:
-```
-
-```{eval-rst}
-.. automodule:: rl4co.models.zoo.symnco.losses
- :members:
- :undoc-members:
-```
-
----
-
-# NonAutoregressive
-
-
-## DeepACO
-
-```{eval-rst}
-.. automodule:: rl4co.models.zoo.deepaco.antsystem
- :members:
- :undoc-members:
-```
-
-```{eval-rst}
-.. automodule:: rl4co.models.zoo.deepaco.model
- :members:
- :undoc-members:
-```
-
-```{eval-rst}
-.. automodule:: rl4co.models.zoo.deepaco.policy
- :members:
- :undoc-members:
-```
-
-```{eval-rst}
-.. automodule:: rl4co.models.zoo.deepaco.decoder
- :members:
- :undoc-members:
-```
-
-
-## NAR-GNN
-
-```{eval-rst}
-.. automodule:: rl4co.models.zoo.nargnn.policy
- :members:
- :undoc-members:
-```
-
-```{eval-rst}
-.. automodule:: rl4co.models.zoo.nargnn.encoder
- :members:
- :undoc-members:
-```
\ No newline at end of file
diff --git a/docs/_content/api/models/zoo/improvement.md b/docs/_content/api/models/zoo/improvement.md
deleted file mode 100644
index eda6ca2c..00000000
--- a/docs/_content/api/models/zoo/improvement.md
+++ /dev/null
@@ -1,5 +0,0 @@
-# Improvement Methods
-
-These methods are trained to improve existing solutions iteratively, akin to local search algorithms. They focus on refining existing solutions rather than generating them from scratch.
-
-_coming soon!_
\ No newline at end of file
diff --git a/docs/_content/api/models/zoo/transductive.md b/docs/_content/api/models/zoo/transductive.md
deleted file mode 100644
index b4e0bbbb..00000000
--- a/docs/_content/api/models/zoo/transductive.md
+++ /dev/null
@@ -1,31 +0,0 @@
-# Transductive Methods
-
-These methods update policy parameters during online testing to improve the solutions of a specific instance.
-
-## Active Search (AS)
-
-```{eval-rst}
-.. automodule:: rl4co.models.zoo.active_search.search
- :members:
- :undoc-members:
-```
-
-## Efficent Active Search (EAS)
-
-```{eval-rst}
-.. automodule:: rl4co.models.zoo.eas.search
- :members:
- :undoc-members:
-```
-
-```{eval-rst}
-.. automodule:: rl4co.models.zoo.eas.decoder
- :members:
- :undoc-members:
-```
-
-```{eval-rst}
-.. automodule:: rl4co.models.zoo.eas.nn
- :members:
- :undoc-members:
-```
\ No newline at end of file
diff --git a/docs/_content/api/tasks.md b/docs/_content/api/tasks.md
deleted file mode 100644
index 53bb82a0..00000000
--- a/docs/_content/api/tasks.md
+++ /dev/null
@@ -1,17 +0,0 @@
-# Tasks: Train and Evaluate
-
-## Train
-
-```{eval-rst}
-.. automodule:: rl4co.tasks.train
- :members:
- :undoc-members:
-```
-
-## Evaluate
-
-```{eval-rst}
-.. automodule:: rl4co.tasks.eval
- :members:
- :undoc-members:
-```
\ No newline at end of file
diff --git a/docs/_content/general/faq.md b/docs/_content/general/faq.md
deleted file mode 100644
index e16ceaca..00000000
--- a/docs/_content/general/faq.md
+++ /dev/null
@@ -1,4 +0,0 @@
-# FAQ
-
-You can submit your questions via GitHub Issues. If asked more than a few times, a question will have its answer added here. If you are looking for collaborations, we would be happy to hear from you on [Slack](https://join.slack.com/t/rl4co/shared_invite/zt-1ytz2c1v4-0IkQ8NQH4TRXIX8PrRmDhQ) π
-
diff --git a/docs/_theme/rl4co/__init__.py b/docs/_theme/rl4co/__init__.py
deleted file mode 100644
index 31492d0e..00000000
--- a/docs/_theme/rl4co/__init__.py
+++ /dev/null
@@ -1,19 +0,0 @@
-"""Pytorch Lightning Sphinx theme.
-
-From https://github.com/shiftlab/lightning_sphinx_theme.
-"""
-from os import path
-
-__version__ = "0.0.31"
-__version_full__ = __version__
-
-
-def get_html_theme_path():
- """Return list of HTML theme paths."""
- cur_dir = path.abspath(path.dirname(path.dirname(__file__)))
- return cur_dir
-
-
-# See http://www.sphinx-doc.org/en/stable/theming.html#distribute-your-theme-as-a-python-package
-def setup(app):
- app.add_html_theme("ncobench_theme", path.abspath(path.dirname(__file__)))
diff --git a/docs/_theme/rl4co/breadcrumbs.html b/docs/_theme/rl4co/breadcrumbs.html
deleted file mode 100644
index e60a62e1..00000000
--- a/docs/_theme/rl4co/breadcrumbs.html
+++ /dev/null
@@ -1,90 +0,0 @@
-{# Support for Sphinx 1.3+ page_source_suffix, but don't break old builds. #}
-
-{% if page_source_suffix %}
-{% set suffix = page_source_suffix %}
-{% else %}
-{% set suffix = source_suffix %}
-{% endif %}
-
-{% if meta is defined and meta is not none %}
-{% set check_meta = True %}
-{% else %}
-{% set check_meta = False %}
-{% endif %}
-
-{% if check_meta and 'github_url' in meta %}
-{% set display_github = True %}
-{% endif %}
-
-{% if check_meta and 'bitbucket_url' in meta %}
-{% set display_bitbucket = True %}
-{% endif %}
-
-{% if check_meta and 'gitlab_url' in meta %}
-{% set display_gitlab = True %}
-{% endif %}
-
-
Don't worry, RL4CO doesn't use your cookie. But maybe ReadTheDoc will have some cookie policy, you could refer it here.
-
-
-
diff --git a/docs/_theme/rl4co/extensions/lightning.py b/docs/_theme/rl4co/extensions/lightning.py
deleted file mode 100644
index 8cba7fda..00000000
--- a/docs/_theme/rl4co/extensions/lightning.py
+++ /dev/null
@@ -1,179 +0,0 @@
-# Copyright The PyTorch Lightning team.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-# http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-from docutils import nodes
-from docutils.statemachine import StringList
-from pt_lightning_sphinx_theme.extensions.pytorch_tutorials import (
- CustomCalloutItemDirective,
- CustomCardItemDirective,
- DisplayItemDirective,
- LikeButtonWithTitle,
- ReactGreeter,
- SlackButton,
- TwoColumns,
- cardnode,
-)
-from sphinx.util.docutils import SphinxDirective
-
-
-class tutoriallistnode(nodes.General, nodes.Element):
- """A placeholder node that we can use during the first parse to later
- replace with the card list."""
-
- pass
-
-
-def visit_cardnode(self, node):
- """Hook to make ``cardnode`` behave the same as a ``paragraph`` node (e.g.
- insert ``
`` tags)."""
- self.visit_paragraph(node)
-
-
-def depart_cardnode(self, node):
- """Hook to make ``cardnode`` behave the same as a ``paragraph`` node (e.g.
- insert ``
`` tags)."""
- self.depart_paragraph(node)
-
-
-def purge_cards(app, env, docname):
- """Hook to purge the cards from the env for the given docname.
-
- This will be called for each changed doc when rebuilding, so will
- reset the card from that doc.
- """
- if not hasattr(env, "all_cardnodes"):
- return
-
- env.all_cardnodes = [node for node in env.all_cardnodes if node["docname"] != docname]
-
-
-def merge_cards(app, env, docnames, other):
- """Hook to merge cards from multiple environments (e.g. when multi-
- threading)."""
- if not hasattr(env, "all_cardnodes"):
- env.all_cardnodes = []
- if hasattr(other, "all_cardnodes"):
- env.all_cardnodes.extend(other.all_cardnodes)
-
-
-TUTORIAL_LIST_START = """
-.. raw:: html
-
-
-
-
-
-
-
-
-
-
-
-
-.. Tutorial cards below this line
-
-"""
-
-TUTORIAL_LIST_END = """
-
-.. End of tutorial card section
-
-.. raw:: html
-
-
-
-
-
-
-
-
-
-
-"""
-
-
-class TutorialListDirective(SphinxDirective):
- """Our custom directive which inserts the header and footer markup for the
- tutorial block with a placeholder node (``tutoriallistnode``) which can be
- modified inplace later with the list of cards."""
-
- def run(self):
- start_list = StringList(TUTORIAL_LIST_START.split("\n"))
- start_node = nodes.paragraph()
- self.state.nested_parse(start_list, self.content_offset, start_node)
-
- end_list = StringList(TUTORIAL_LIST_END.split("\n"))
- end_node = nodes.paragraph()
- self.state.nested_parse(end_list, self.content_offset, end_node)
-
- return start_node.children + [tutoriallistnode("")] + end_node.children
-
-
-def process_card_nodes(app, doctree, fromdocname):
- """This hook does two things.
-
- 1. Find any ``cardnode`` nodes in the document and remove them (we don't want them to render on the page).
- 2. Find any ``tutoriallistnode`` nodes in the document and inplace edit them to have the list of ``cardnode`` nodes
- from the environment.
- """
- for card in doctree.traverse(cardnode):
- card.parent.remove(card)
-
- for node in doctree.traverse(tutoriallistnode):
- content = []
-
- for card in app.builder.env.all_cardnodes:
- content.append(card["node"])
-
- node.replace_self(content)
-
-
-def setup(app):
- """Set-up the extension.
-
- Add out custom nodes and directives, then attach our hooks to the
- required events.
- """
- app.add_node(tutoriallistnode)
- app.add_node(
- cardnode,
- html=(visit_cardnode, depart_cardnode),
- latex=(visit_cardnode, depart_cardnode),
- text=(visit_cardnode, depart_cardnode),
- )
-
- app.add_directive("customcarditem", CustomCardItemDirective)
- app.add_directive("displayitem", DisplayItemDirective)
- app.add_directive("join_slack", SlackButton)
- app.add_directive("like_button_with_title", LikeButtonWithTitle)
- app.add_directive("react_greeter", ReactGreeter)
- app.add_directive("twocolumns", TwoColumns)
- app.add_directive("customcalloutitem", CustomCalloutItemDirective)
- app.add_directive("tutoriallist", TutorialListDirective)
-
- app.connect("env-purge-doc", purge_cards)
- app.connect("env-merge-info", merge_cards)
- app.connect("doctree-resolved", process_card_nodes)
-
- return {
- "parallel_read_safe": True,
- "parallel_write_safe": True,
- }
diff --git a/docs/_theme/rl4co/extensions/pytorch_tutorials.py b/docs/_theme/rl4co/extensions/pytorch_tutorials.py
deleted file mode 100644
index 12b0e73d..00000000
--- a/docs/_theme/rl4co/extensions/pytorch_tutorials.py
+++ /dev/null
@@ -1,523 +0,0 @@
-"""Adapted from PyTorch Tutorials: https://github.com/pytorch/tutorials.
-
-BSD 3-Clause License
-
-Copyright (c) 2017, Pytorch contributors
-All rights reserved.
-
-Redistribution and use in source and binary forms, with or without
-modification, are permitted provided that the following conditions are met:
-
-* Redistributions of source code must retain the above copyright notice, this
- list of conditions and the following disclaimer.
-
-* Redistributions in binary form must reproduce the above copyright notice,
- this list of conditions and the following disclaimer in the documentation
- and/or other materials provided with the distribution.
-
-* Neither the name of the copyright holder nor the names of its
- contributors may be used to endorse or promote products derived from
- this software without specific prior written permission.
-
-THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
-AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
-IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
-DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
-FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
-DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
-SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
-CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
-OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-"""
-
-from docutils import nodes
-from docutils.parsers.rst import Directive, directives
-from docutils.statemachine import StringList
-from pt_lightning_sphinx_theme.extensions.react import get_react_component_rst
-from sphinx.util.docutils import SphinxDirective
-
-try:
- FileNotFoundError
-except NameError:
- FileNotFoundError = IOError
-
-
-class cardnode(nodes.General, nodes.TextElement):
- pass
-
-
-class CustomCardItemDirective(SphinxDirective):
- option_spec = {
- "header": directives.unchanged,
- "image": directives.unchanged,
- "card_description": directives.unchanged,
- "tags": directives.unchanged,
- "beta": directives.flag,
- }
-
- def run(self):
- try:
- if "header" in self.options:
- header = self.options["header"]
- else:
- raise ValueError("header not doc found")
-
- if "image" in self.options:
- image = ""
- else:
- image = ""
-
- # TODO: This probably only works when the tutorial list directive is in index.html
- link = self.env.docname + ".html"
-
- if "card_description" in self.options:
- card_description = self.options["card_description"]
- else:
- card_description = ""
-
- if "tags" in self.options:
- tags = self.options["tags"]
- else:
- tags = ""
-
- if "beta" in self.options:
- beta = "Beta"
- else:
- beta = ""
-
- except FileNotFoundError as e:
- print(e)
- return []
- except ValueError as e:
- print(e)
- raise
- return []
-
- card_rst = CARD_TEMPLATE.format(
- header=header,
- image=image,
- link=link,
- card_description=card_description,
- tags=tags,
- beta=beta,
- )
- card_list = StringList(card_rst.split("\n"))
- node = cardnode()
- self.state.nested_parse(card_list, self.content_offset, node)
-
- if not hasattr(self.env, "all_cardnodes"):
- self.env.all_cardnodes = []
- self.env.all_cardnodes.append({"docname": self.env.docname, "node": node})
- return [node]
-
-
-CARD_TEMPLATE = """
-.. raw:: html
-
-
-
- {% trans %}Free document hosting provided by Read the Docs.{% endtrans %}
-
-
-
-{% endif %}
diff --git a/docs/assets/figs/embeddings.png b/docs/assets/figs/embeddings.png
new file mode 100644
index 00000000..bbfc3325
Binary files /dev/null and b/docs/assets/figs/embeddings.png differ
diff --git a/docs/assets/figs/framework.png b/docs/assets/figs/framework.png
new file mode 100644
index 00000000..b4750966
Binary files /dev/null and b/docs/assets/figs/framework.png differ
diff --git a/docs/assets/figs/policies.png b/docs/assets/figs/policies.png
new file mode 100644
index 00000000..862f9eb9
Binary files /dev/null and b/docs/assets/figs/policies.png differ
diff --git a/docs/assets/figs/rl4co-logo.svg b/docs/assets/figs/rl4co-logo.svg
new file mode 100644
index 00000000..041aa2e1
--- /dev/null
+++ b/docs/assets/figs/rl4co-logo.svg
@@ -0,0 +1,38 @@
+
+
+
diff --git a/docs/assets/rl4co_animated_full_tricks.svg b/docs/assets/rl4co_animated_full_tricks.svg
new file mode 100644
index 00000000..2e1ac249
--- /dev/null
+++ b/docs/assets/rl4co_animated_full_tricks.svg
@@ -0,0 +1,77 @@
+
+
+
diff --git a/docs/conf.py b/docs/conf.py
deleted file mode 100644
index fca346f6..00000000
--- a/docs/conf.py
+++ /dev/null
@@ -1,121 +0,0 @@
-# Configuration file for the Sphinx documentation builder.
-#
-# For the full list of built-in configuration values, see the documentation:
-# https://www.sphinx-doc.org/en/master/usage/configuration.html
-
-# -- Project information -----------------------------------------------------
-# https://www.sphinx-doc.org/en/master/usage/configuration.html#project-information
-import rl4co
-
-project = "rl4co"
-copyright = "RL4CO Contributors"
-author = "Federico Berto, Chuanbo Hua, Junyoung Park, Minsu Kim, Hyeonah Kim, Jiwoo Son, Haeyeon Kim, Joungho Kim, Jinkyoo Park"
-
-
-release = rl4co.__version__
-
-# -- General configuration ---------------------------------------------------
-# https://www.sphinx-doc.org/en/master/usage/configuration.html#general-configuration
-
-
-# Most taken from: https://github.com/Lightning-AI/lightning/blob/master/docs/source-pytorch/conf.py
-extensions = [
- "sphinx.ext.autodoc",
- "sphinx.ext.doctest",
- "sphinx.ext.intersphinx",
- "sphinx_toolbox.collapse",
- "sphinx.ext.todo",
- "sphinx.ext.coverage",
- "sphinx.ext.viewcode",
- "sphinx.ext.autosummary",
- "sphinx.ext.napoleon",
- "sphinx.ext.autosectionlabel",
- "sphinxcontrib.video",
- "sphinxcontrib.katex",
- "myst_parser",
- "nbsphinx",
- "sphinx_autodoc_typehints",
- "sphinx_copybutton",
- "sphinx_paramlinks",
- "sphinx_togglebutton",
- "sphinxcontrib.collections",
-]
-
-
-# Include the folder from the main repo containing the notebooks
-collections = {
- "my_files": {
- "driver": "copy_folder",
- "source": "../examples/",
- "target": "",
- "ignore": ["*.ckpt"],
- }
-}
-
-templates_path = ["_templates"]
-exclude_patterns = ["_build", "Thumbs.db", ".DS_Store"]
-
-# -- Options for HTML output -------------------------------------------------
-# https://www.sphinx-doc.org/en/master/usage/configuration.html#options-for-html-output
-
-# SECTION: original theme set
-# html_theme = 'alabaster'
-# html_static_path = ['_static']
-
-# SECTION: set sphinx_rtd_theme
-# html_theme = "sphinx_rtd_theme"
-# html_static_path = ['_static']
-# html_theme_options = {
-# 'navigation_depth': 4,
-# }
-
-# SECTION: set rl4co theme
-html_theme = "rl4co"
-html_theme_path = ["_theme"]
-
-# SECTION: set different parser for rst and md
-source_suffix = {
- ".rst": "restructuredtext",
- ".md": "markdown",
-}
-
-# intersphinx_mapping = {
-# "python": ("https://docs.python.org/3", None),
-# "torch": ("https://pytorch.org/docs/stable/", None),
-# "torchmetrics": ("https://torchmetrics.readthedocs.io/en/stable/", None),
-# "tensordict": ("https://pytorch-labs.github.io/tensordict/", None),
-# "torchrl": ("https://pytorch.org/rl/", None),
-# "torchaudio": ("https://pytorch.org/audio/stable/", None),
-# "torchtext": ("https://pytorch.org/text/stable/", None),
-# "torchvision": ("https://pytorch.org/vision/stable/", None),
-# "numpy": ("https://numpy.org/doc/stable/", None),
-# }
-
-
-autosummary_generate = True
-
-autodoc_member_order = "groupwise"
-
-autoclass_content = "both"
-
-autodoc_default_options = {
- "members": True,
- "methods": True,
- "special-members": "__call__",
- "exclude-members": "_abc_impl",
- "show-inheritance": True,
-}
-
-# Sphinx will add βpermalinksβ for each heading and description environment as paragraph signs that
-# become visible when the mouse hovers over them.
-# This value determines the text for the permalink; it defaults to "ΒΆ". Set it to None or the empty
-# string to disable permalinks.
-# https://www.sphinx-doc.org/en/master/usage/configuration.html#confval-html_add_permalinks
-html_permalinks = True
-html_permalinks_icon = "ΒΆ"
-
-# True to prefix each section label with the name of the document it is in, followed by a colon.
-# For example, index:Introduction for a section called Introduction that appears in document index.rst.
-# Useful for avoiding ambiguity when the same section heading appears in different documents.
-# http://www.sphinx-doc.org/en/master/usage/extensions/autosectionlabel.html
-autosectionlabel_prefix_document = True
diff --git a/docs/content/api/data.md b/docs/content/api/data.md
new file mode 100644
index 00000000..b316c783
--- /dev/null
+++ b/docs/content/api/data.md
@@ -0,0 +1,23 @@
+## Datasets
+
+:::data.dataset
+ options:
+ show_root_heading: false
+
+## Data Generation
+
+:::data.generate_data
+ options:
+ show_root_heading: false
+
+## Transforms
+
+:::data.transforms
+ options:
+ show_root_heading: false
+
+## Utils
+
+:::data.utils
+ options:
+ show_root_heading: false
\ No newline at end of file
diff --git a/docs/content/api/decoding.md b/docs/content/api/decoding.md
new file mode 100644
index 00000000..0a0b5ec0
--- /dev/null
+++ b/docs/content/api/decoding.md
@@ -0,0 +1,5 @@
+# Decoding Strategies
+
+:::utils.decoding
+ options:
+ show_root_heading: false
diff --git a/docs/_content/api/envs/base.md b/docs/content/api/envs/base.md
similarity index 50%
rename from docs/_content/api/envs/base.md
rename to docs/content/api/envs/base.md
index 30e1a190..fe914cb6 100644
--- a/docs/_content/api/envs/base.md
+++ b/docs/content/api/envs/base.md
@@ -1,21 +1,15 @@
# Base Environment
-## `RL4COEnvBase`
-
This is the base wrapper around TorchRL's `EnvBase`, with additional functionality.
-```{eval-rst}
-.. automodule:: rl4co.envs.common.base
- :members:
- :undoc-members:
-```
+::: envs.common.base.RL4COEnvBase
+::: envs.common.base.ImprovementEnvBase
+
## Utilities
These contain utilities such as the base `Generator` class and `get_sampler`.
-```{eval-rst}
-.. automodule:: rl4co.envs.common.utils
- :members:
- :undoc-members:
-```
\ No newline at end of file
+:::envs.common.utils.Generator
+:::envs.common.utils.get_sampler
+:::envs.common.utils.batch_to_scalar
\ No newline at end of file
diff --git a/docs/content/api/envs/eda.md b/docs/content/api/envs/eda.md
new file mode 100644
index 00000000..c070e0f4
--- /dev/null
+++ b/docs/content/api/envs/eda.md
@@ -0,0 +1,14 @@
+# EDA Problems
+Environment for Electronic Design Automation (EDA) problems
+
+## Decap Placement Problem (DPP)
+
+:::envs.eda.dpp.env.DPPEnv
+
+:::envs.eda.dpp.generator.DPPGenerator
+
+## Multi-port Decap Placement Problem (mDPP)
+
+:::envs.eda.mdpp.env.MDPPEnv
+
+:::envs.eda.mdpp.generator.MDPPGenerator
diff --git a/docs/content/api/envs/routing.md b/docs/content/api/envs/routing.md
new file mode 100644
index 00000000..7c5be49f
--- /dev/null
+++ b/docs/content/api/envs/routing.md
@@ -0,0 +1,61 @@
+# Routing Problems
+
+See also the [Multi-Task VRP](#Multi-Task-Vehicle-Routing-Problem-(MTVRP)) at the bottom of this page, that includes 16 variants!
+
+## Asymmetric Traveling Salesman Problem (ATSP)
+
+:::envs.routing.atsp.env.ATSPEnv
+
+:::envs.routing.atsp.generator.ATSPGenerator
+
+## Capacitated Vehicle Routing Problem (CVRP)
+
+:::envs.routing.cvrp.env.CVRPEnv
+
+:::envs.routing.cvrp.generator.CVRPGenerator
+
+## Multiple Traveling Salesman Problem (mTSP)
+
+:::envs.routing.mtsp.env.MTSPEnv
+
+:::envs.routing.mtsp.generator.MTSPGenerator
+
+
+## Orienteering Problem (OP)
+
+:::envs.routing.op.env.OPEnv
+
+:::envs.routing.op.generator.OPGenerator
+
+## Pickup and Delivery Problem (PDP)
+
+:::envs.routing.pdp.env.PDPEnv
+
+:::envs.routing.pdp.generator.PDPGenerator
+
+## Prize Collecting Traveling Salesman Problem (PCTSP)
+
+:::envs.routing.pctsp.env.PCTSPEnv
+
+:::envs.routing.pctsp.generator.PCTSPGenerator
+
+## Split Delivery Vehicle Routing Problem (SDVRP)
+
+:::envs.routing.sdvrp.env.SDVRPEnv
+
+## Stochastic Prize Collecting Traveling Salesman Problem (SPCTSP)
+
+:::envs.routing.spctsp.env.SPCTSPEnv
+
+## Traveling Salesman Problem (TSP)
+
+:::envs.routing.tsp.env.TSPEnv
+
+:::envs.routing.tsp.generator.TSPGenerator
+
+## Multi-Task Vehicle Routing Problem (MTVRP)
+
+
+:::envs.routing.mtvrp.env.MTVRPEnv
+
+:::envs.routing.mtvrp.generator.MTVRPGenerator
\ No newline at end of file
diff --git a/docs/content/api/envs/scheduling.md b/docs/content/api/envs/scheduling.md
new file mode 100644
index 00000000..62b172b6
--- /dev/null
+++ b/docs/content/api/envs/scheduling.md
@@ -0,0 +1,13 @@
+# Scheduling Problems
+
+## Flexible Flow Shop Problem (FFSP)
+
+:::envs.scheduling.ffsp.env.FFSPEnv
+
+:::envs.scheduling.ffsp.generator.FFSPGenerator
+
+## Single Machine Total Weighted Tardiness Problem (SMTWTP)
+
+:::envs.scheduling.smtwtp.env.SMTWTPEnv
+
+:::envs.scheduling.smtwtp.generator.SMTWTPGenerator
\ No newline at end of file
diff --git a/docs/content/api/networks/base_policies.md b/docs/content/api/networks/base_policies.md
new file mode 100644
index 00000000..0d895afb
--- /dev/null
+++ b/docs/content/api/networks/base_policies.md
@@ -0,0 +1,42 @@
+
+# Constructive Policies Base Classes
+
+:::models.common.constructive.base
+ options:
+ show_root_heading: false
+
+## Autoregressive Policies
+
+:::models.common.constructive.autoregressive.encoder
+ options:
+ show_root_heading: false
+
+:::models.common.constructive.autoregressive.decoder
+ options:
+ show_root_heading: false
+
+:::models.common.constructive.autoregressive.policy
+ options:
+ show_root_heading: false
+
+## Nonautoregressive Policies
+
+:::models.common.constructive.nonautoregressive.encoder
+ options:
+ show_root_heading: false
+
+:::models.common.constructive.nonautoregressive.decoder
+ options:
+ show_root_heading: false
+
+:::models.common.constructive.nonautoregressive.policy
+ options:
+ show_root_heading: false
+
+
+# Improvement Policies (Base Classes)
+
+
+:::models.common.improvement.base
+ options:
+ show_root_heading: false
\ No newline at end of file
diff --git a/docs/_content/api/models/env_embeddings.md b/docs/content/api/networks/env_embeddings.md
similarity index 79%
rename from docs/_content/api/models/env_embeddings.md
rename to docs/content/api/networks/env_embeddings.md
index 9d70ae6b..5056cb3b 100644
--- a/docs/_content/api/models/env_embeddings.md
+++ b/docs/content/api/networks/env_embeddings.md
@@ -9,39 +9,26 @@ In autoregressive policies, environment embeddings transfer data from feature sp
-
-
## Context Embeddings
The context embedding is used to modify the query embedding of the problem node of the current partial solution. Usually consists of a projection of gathered node embeddings and features to the embedding space.
-
-```{eval-rst}
-.. automodule:: rl4co.models.nn.env_embeddings.context
- :members:
- :undoc-members:
-```
-
----
+:::models.nn.env_embeddings.context
+ options:
+ show_root_heading: false
## Dynamic Embeddings
The dynamic embedding is used to modify query, key and value vectors of the attention mechanism based on the current state of the environment (which is changing during the rollout). Generally consists of a linear layer that projects the node features to the embedding space.
-```{eval-rst}
-.. automodule:: rl4co.models.nn.env_embeddings.dynamic
- :members:
- :undoc-members:
-```
-
----
+:::models.nn.env_embeddings.dynamic
+ options:
+ show_root_heading: false
## Init Embeddings
The init embedding is used to initialize the general embedding of the problem nodes without any solution information. Generally consists of a linear layer that projects the node features to the embedding space.
-```{eval-rst}
-.. automodule:: rl4co.models.nn.env_embeddings.init
- :members:
- :undoc-members:
-```
\ No newline at end of file
+:::models.nn.env_embeddings.init
+ options:
+ show_root_heading: false
\ No newline at end of file
diff --git a/docs/content/api/networks/improvement_policies.md b/docs/content/api/networks/improvement_policies.md
new file mode 100644
index 00000000..a4fdc075
--- /dev/null
+++ b/docs/content/api/networks/improvement_policies.md
@@ -0,0 +1,7 @@
+## Improvement Policies (Base Classes)
+
+
+
+:::models.common.improvement.base
+ options:
+ show_root_heading: false
\ No newline at end of file
diff --git a/docs/content/api/networks/nn.md b/docs/content/api/networks/nn.md
new file mode 100644
index 00000000..66cc5637
--- /dev/null
+++ b/docs/content/api/networks/nn.md
@@ -0,0 +1,29 @@
+# Neural Network Modules
+
+## Critic Network
+
+:::models.rl.common.critic.CriticNetwork
+
+## Graph Neural Networks
+
+:::models.nn.graph.attnnet.MultiHeadAttentionLayer
+:::models.nn.graph.attnnet.GraphAttentionNetwork
+
+:::models.nn.graph.gcn.GCNEncoder
+
+:::models.nn.graph.mpnn.MessagePassingEncoder
+
+## Attention Mechanisms
+
+:::models.nn.attention
+ options:
+ show_root_heading: false
+
+## Multi-Layer Perceptron
+
+:::models.nn.mlp.MLP
+
+## Operations
+
+:::models.nn.ops.PositionalEncoding
+:::models.nn.ops.RandomEncoding
\ No newline at end of file
diff --git a/docs/content/api/rl/a2c.md b/docs/content/api/rl/a2c.md
new file mode 100644
index 00000000..77abe89f
--- /dev/null
+++ b/docs/content/api/rl/a2c.md
@@ -0,0 +1 @@
+:::models.rl.a2c.a2c.A2C
diff --git a/docs/content/api/rl/base.md b/docs/content/api/rl/base.md
new file mode 100644
index 00000000..58ebf0aa
--- /dev/null
+++ b/docs/content/api/rl/base.md
@@ -0,0 +1,19 @@
+# RL4COLitModule
+
+The `RL4COLitModule` is a wrapper around PyTorch Lightning's `LightningModule` that provides additional functionality for RL algorithms. It is the parent class for all RL algorithms in the library.
+
+::: models.rl.common.base.RL4COLitModule
+
+
+## Transductive Learning
+
+Transductive models are learning algorithms that optimize on a specific instance. They improve solutions by updating policy parameters $\theta$, which means that we are running optimization (backprop) **at test time**. Transductive learning can be performed with different policies: for example EAS updates (a part of) AR policies parameters to obtain better solutions, but I guess there are ways (or papers out there I don't know of) that optimize at test time.
+
+
+!!! tip
+ You may refer to the definition of [inductive vs transductive RL](https://en.wikipedia.org/wiki/Transduction_(machine_learning)) . In inductive RL, we train to generalize to new instances. In transductive RL we train (or finetune) to solve only specific ones.
+
+
+:::models.common.transductive.base
+ options:
+ show_root_heading: false
\ No newline at end of file
diff --git a/docs/content/api/rl/ppo.md b/docs/content/api/rl/ppo.md
new file mode 100644
index 00000000..04c9a4bc
--- /dev/null
+++ b/docs/content/api/rl/ppo.md
@@ -0,0 +1 @@
+:::models.rl.ppo.ppo.PPO
\ No newline at end of file
diff --git a/docs/content/api/rl/reinforce.md b/docs/content/api/rl/reinforce.md
new file mode 100644
index 00000000..60e25ae7
--- /dev/null
+++ b/docs/content/api/rl/reinforce.md
@@ -0,0 +1,10 @@
+:::models.rl.reinforce.reinforce.REINFORCE
+:::models.rl.reinforce.baselines.REINFORCEBaseline
+:::models.rl.reinforce.baselines.NoBaseline
+:::models.rl.reinforce.baselines.SharedBaseline
+:::models.rl.reinforce.baselines.ExponentialBaseline
+:::models.rl.reinforce.baselines.MeanBaseline
+:::models.rl.reinforce.baselines.WarmupBaseline
+:::models.rl.reinforce.baselines.CriticBaseline
+:::models.rl.reinforce.baselines.RolloutBaseline
+:::models.rl.reinforce.baselines.get_reinforce_baseline
diff --git a/docs/content/api/tasks.md b/docs/content/api/tasks.md
new file mode 100644
index 00000000..37e0ac7e
--- /dev/null
+++ b/docs/content/api/tasks.md
@@ -0,0 +1,9 @@
+## Train
+:::tasks.train
+ options:
+ show_root_heading: false
+
+## Evaluate
+:::tasks.eval
+ options:
+ show_root_heading: false
\ No newline at end of file
diff --git a/docs/_theme/rl4co/extensions/__init__.py b/docs/content/api/train_and_eval.md
similarity index 100%
rename from docs/_theme/rl4co/extensions/__init__.py
rename to docs/content/api/train_and_eval.md
diff --git a/docs/content/api/zoo/constructive_ar.md b/docs/content/api/zoo/constructive_ar.md
new file mode 100644
index 00000000..5391eeb1
--- /dev/null
+++ b/docs/content/api/zoo/constructive_ar.md
@@ -0,0 +1,113 @@
+# Constructive Autoregressive Methods
+
+## Attention Model (AM)
+
+:::models.zoo.am.model
+ options:
+ show_root_heading: false
+
+:::models.zoo.am.policy
+ options:
+ show_root_heading: false
+
+## Attention Model - PPO (AM-PPO)
+
+:::models.zoo.amppo.model
+ options:
+ show_root_heading: false
+
+## Heterogeneous Attention Model (HAM)
+
+:::models.zoo.ham.model
+ options:
+ show_root_heading: false
+
+:::models.zoo.ham.policy
+ options:
+ show_root_heading: false
+
+:::models.zoo.ham.encoder
+ options:
+ show_root_heading: false
+
+:::models.zoo.ham.attention
+ options:
+ show_root_heading: false
+
+## Matrix Encoding Network (MatNet)
+
+:::models.zoo.matnet.model
+ options:
+ show_root_heading: false
+
+:::models.zoo.matnet.policy
+ options:
+ show_root_heading: false
+
+:::models.zoo.matnet.encoder
+ options:
+ show_root_heading: false
+
+:::models.zoo.matnet.decoder
+ options:
+ show_root_heading: false
+
+## Multi-Decoder Attention Model (MDAM)
+
+:::models.zoo.mdam.model
+ options:
+ show_root_heading: false
+
+:::models.zoo.mdam.policy
+ options:
+ show_root_heading: false
+
+:::models.zoo.mdam.encoder
+ options:
+ show_root_heading: false
+
+:::models.zoo.mdam.decoder
+ options:
+ show_root_heading: false
+
+## POMO
+
+:::models.zoo.pomo.model
+ options:
+ show_root_heading: false
+
+## Pointer Network (PtrNet)
+
+:::models.zoo.ptrnet.model
+ options:
+ show_root_heading: false
+
+:::models.zoo.ptrnet.policy
+ options:
+ show_root_heading: false
+
+:::models.zoo.ptrnet.encoder
+ options:
+ show_root_heading: false
+
+:::models.zoo.ptrnet.decoder
+ options:
+ show_root_heading: false
+
+:::models.zoo.ptrnet.critic
+ options:
+ show_root_heading: false
+
+## SymNCO
+
+:::models.zoo.symnco.model
+ options:
+ show_root_heading: false
+
+:::models.zoo.symnco.policy
+ options:
+ show_root_heading: false
+
+:::models.zoo.symnco.losses
+ options:
+ show_root_heading: false
diff --git a/docs/content/api/zoo/constructive_nar.md b/docs/content/api/zoo/constructive_nar.md
new file mode 100644
index 00000000..3e1860ec
--- /dev/null
+++ b/docs/content/api/zoo/constructive_nar.md
@@ -0,0 +1,27 @@
+
+# Constructive NonAutoregressive
+
+## DeepACO
+
+:::models.zoo.deepaco.antsystem
+ options:
+ show_root_heading: false
+
+:::models.zoo.deepaco.model
+ options:
+ show_root_heading: false
+
+:::models.zoo.deepaco.policy
+ options:
+ show_root_heading: false
+
+
+## NAR-GNN
+
+:::models.zoo.nargnn.policy
+ options:
+ show_root_heading: false
+
+:::models.zoo.nargnn.encoder
+ options:
+ show_root_heading: false
diff --git a/docs/content/api/zoo/improvement.md b/docs/content/api/zoo/improvement.md
new file mode 100644
index 00000000..8ca68c75
--- /dev/null
+++ b/docs/content/api/zoo/improvement.md
@@ -0,0 +1,55 @@
+# Improvement Methods
+
+These methods are trained to improve existing solutions iteratively, akin to local search algorithms. They focus on refining existing solutions rather than generating them from scratch.
+
+### DACT
+
+:::models.zoo.dact.encoder
+ options:
+ show_root_heading: false
+
+:::models.zoo.dact.decoder
+ options:
+ show_root_heading: false
+
+:::models.zoo.dact.policy
+ options:
+ show_root_heading: false
+
+:::models.zoo.dact.model
+ options:
+ show_root_heading: false
+
+
+### N2S
+
+:::models.zoo.n2s.encoder
+ options:
+ show_root_heading: false
+
+:::models.zoo.n2s.decoder
+ options:
+ show_root_heading: false
+
+:::models.zoo.n2s.policy
+ options:
+ show_root_heading: false
+
+:::models.zoo.n2s.model
+ options:
+ show_root_heading: false
+
+
+### NeuOpt
+
+:::models.zoo.neuopt.decoder
+ options:
+ show_root_heading: false
+
+:::models.zoo.neuopt.policy
+ options:
+ show_root_heading: false
+
+:::models.zoo.neuopt.model
+ options:
+ show_root_heading: false
\ No newline at end of file
diff --git a/docs/content/api/zoo/transductive.md b/docs/content/api/zoo/transductive.md
new file mode 100644
index 00000000..1b7281ad
--- /dev/null
+++ b/docs/content/api/zoo/transductive.md
@@ -0,0 +1,23 @@
+## Transductive Methods
+
+These methods update policy parameters during online testing to improve the solutions of a specific instance.
+
+### Active Search (AS)
+
+:::models.zoo.active_search.search
+ options:
+ show_root_heading: false
+
+### Efficent Active Search (EAS)
+
+:::models.zoo.eas.search
+ options:
+ show_root_heading: false
+
+:::models.zoo.eas.decoder
+ options:
+ show_root_heading: false
+
+:::models.zoo.eas.nn
+ options:
+ show_root_heading: false
diff --git a/docs/content/environments/eda.md b/docs/content/environments/eda.md
new file mode 100644
index 00000000..e69de29b
diff --git a/docs/content/environments/routing.md b/docs/content/environments/routing.md
new file mode 100644
index 00000000..e69de29b
diff --git a/docs/content/environments/scheduling.md b/docs/content/environments/scheduling.md
new file mode 100644
index 00000000..e69de29b
diff --git a/docs/content/general/ai4co.md b/docs/content/general/ai4co.md
new file mode 100644
index 00000000..66db67a3
--- /dev/null
+++ b/docs/content/general/ai4co.md
@@ -0,0 +1,14 @@
+# AI4CO Community
+
+We invite you to join our AI4CO community, an open and inclusive research group in Artificial Intelligence (AI) for Combinatorial Optimization (CO)!
+
+
+## Links
+- [GitHub](https://github.com/ai4co)
+- [Slack](https://bit.ly/ai4co-slack)
+- [Website](https://ai4co.org) (coming soon!)
+
+
+
+
+
\ No newline at end of file
diff --git a/docs/_content/general/contribute.md b/docs/content/general/contribute.md
similarity index 100%
rename from docs/_content/general/contribute.md
rename to docs/content/general/contribute.md
diff --git a/docs/content/general/faq.md b/docs/content/general/faq.md
new file mode 100644
index 00000000..c84cb0f6
--- /dev/null
+++ b/docs/content/general/faq.md
@@ -0,0 +1,11 @@
+# FAQ
+
+
+
+You can submit your questions via [GitHub Issues](https://github.com/ai4co/rl4co/discussions) or [Discussions](https://github.com/ai4co/rl4co/discussions). You may search for your question in the existing issues or discussions before submitting a new one.
+
+
+If ask more than a few times, we will add it here!
+
+
+
diff --git a/docs/content/intro/environments.md b/docs/content/intro/environments.md
new file mode 100644
index 00000000..4de3c555
--- /dev/null
+++ b/docs/content/intro/environments.md
@@ -0,0 +1,28 @@
+## Definition
+
+Given a CO problem instance $\mathbf{x}$, we formulate the solution-generating procedure as a Markov Decision Process (MDP) characterized by a tuple $(\mathcal{S}, \mathcal{A}, \mathcal{T}, \mathcal{R}, \gamma)$ as follows:
+
+- **State** $\mathcal{S}$ is the space of states that represent the given problem $\mathbf{x}$ and the current partial solution being updated in the MDP.
+- **Action** $\mathcal{A}$ is the action space, which includes all feasible actions $a_t$ that can be taken at each step $t$.
+- **State Transition** $\mathcal{T}$ is the deterministic state transition function $s_{t+1} = \mathcal{T}(s_t, a_t)$ that updates a state $s_t$ to the next state $s_{t+1}$.
+- **Reward** $\mathcal{R}$ is the reward function $\mathcal{R}(s_t, a_t)$ representing the immediate reward received after taking action $a_t$ in state $s_t$.
+- **Discount Factor** $\gamma \in [0, 1]$ determines the importance of future rewards.
+
+Since the state transition is deterministic, we represent the solution for a problem $\mathbf{x}$ as a sequence of $T$ actions $\mathbf{a} = (a_1, \ldots, a_T)$. Then the total return $\sum_{t=1}^T \mathcal{R}(s_t, a_t)$ translates to the negative cost function of the CO problem.
+
+## Implementation
+
+Environments in our library fully specify the CO problems and their logic. They are based on the `RL4COEnvBase` class that extends from the [`EnvBase`](https://pytorch.org/rl/stable/reference/generated/torchrl.envs.EnvBase.html#torchrl.envs.EnvBase) in TorchRL.
+
+Key features:
+- A modular `generator` can be provided to the environment.
+- The generator provides CO instances to the environment, and different generators can be used to generate different data distributions.
+- Static instance data and dynamic variables, such as the current state $s_t$, current solution $\mathbf{a}^k$ for improvement environments, policy actions $a_t$, rewards, and additional information are passed in a *stateless* fashion in a [`TensorDict`](https://pytorch.org/tensordict/stable/reference/generated/tensordict.TensorDict.html?highlight=tensordict#tensordict.TensorDict), that we call `td`, through the environment `reset` and `step` functions.
+
+Our environment API contains several functions:
+- `render`
+- `check_solution_validity`
+- `select_start_nodes` (i.e., for POMO-based optimization)
+- Optional API such as `local_search` for solution improvement
+
+It's worth noting that our library enhances the efficiency of environments when compared to vanilla TorchRL, by overriding and optimizing some methods in TorchRL [`EnvBase`](https://pytorch.org/rl/stable/reference/generated/torchrl.envs.EnvBase.html#torchrl.envs.EnvBase). For instance, our new `step` method brings a decrease of up to 50% in latency and halves the memory impact by avoiding saving duplicate components in the stateless [`TensorDict`](https://pytorch.org/tensordict/stable/reference/generated/tensordict.TensorDict.html?highlight=tensordict#tensordict.TensorDict).
\ No newline at end of file
diff --git a/docs/content/intro/intro.md b/docs/content/intro/intro.md
new file mode 100644
index 00000000..ea9f6dd2
--- /dev/null
+++ b/docs/content/intro/intro.md
@@ -0,0 +1,34 @@
+# Introduction
+
+RL4CO is an extensive Reinforcement Learning (RL) for Combinatorial Optimization (CO) benchmark. Our goal is to provide a unified framework for RL-based CO algorithms, and to facilitate reproducible research in this field, decoupling the science from the engineering.
+
+
+
+
+## Motivation
+
+### Why NCO?
+Neural Combinatorial Optimization (NCO) is a subfield of AI that aims to solve combinatorial optimization problems using neural networks. NCO has been successfully applied to a wide range of problems, such as the routing problems in logistics, the scheduling problems in manufacturing, and electronic design automation. The key idea behind NCO is to learn a policy that maps the input data to the optimal solution, without the need for hand-crafted heuristics or domain-specific knowledge.
+
+
+### Why RL?
+Reinforcement Learning (RL) is a machine learning paradigm that enables agents to learn how to make decisions by interacting with an environment. RL has been successfully applied to a wide range of problems, such as playing games, controlling robots, and optimizing complex systems. The key idea behind RL is to learn a policy that maps the state of the environment to the optimal action, by maximizing a reward signal. Importantly, optimal solutions are not required for training, as RL agents learn from the feedback they receive from the environment.
+
+
+
+## Contents
+
+We explore in other pages the following components:
+
+- [Environments](environments.md): Markov Decision Process (MDP) for CO problems and base classes for environments. These are based on [TorchRL](https://pytorch.org/rl/stable/index.html).
+
+- [Policies](policies.md): the neural networks that are used to solve CO problems and their base classes. These are based on [PyTorch](https://pytorch.org/).
+
+- [RL Algorithms](rl.md): (broadly: "models"), which are the processes used to train the policies and their base classes. These are based on [PyTorch Lightning](https://lightning.ai/docs/pytorch/stable/).
+
+
+
+
+
+
+
diff --git a/docs/content/intro/policies.md b/docs/content/intro/policies.md
new file mode 100644
index 00000000..9b5ea4ec
--- /dev/null
+++ b/docs/content/intro/policies.md
@@ -0,0 +1,53 @@
+# Policies
+
+The policies can be categorized into constructive policies, which generate a solution from scratch, and improvement policies, which refine an existing solution.
+
+## Constructive policies
+
+A policy $\pi$ is used to construct a solution from scratch for a given problem instance $\mathbf{x}$. It can be further categorized into autoregressive (AR) and non-autoregressive (NAR) policies.
+
+
+### Autoregressive (AR) policies
+An AR policy is composed of an encoder $f$ that maps the instance $\mathbf{x}$ into an embedding space $\mathbf{h}=f(\mathbf{x})$ and by a decoder $g$ that iteratively determines a sequence of actions $\mathbf{a}$ as follows:
+
+$$
+a_t \sim g(a_t | a_{t-1}, ... ,a_0, s_t, \mathbf{h}), \quad
+\pi(\mathbf{a}|\mathbf{x}) \triangleq \prod_{t=1}^{T-1} g(a_{t} | a_{t-1}, \ldots ,a_0, s_t, \mathbf{h}).
+$$
+
+### Non-autoregressive (NAR) policies
+A NAR policy encodes a problem $\mathbf{x}$ into a heuristic $\mathcal{H} = f(\mathbf{x}) \in \mathbb{R}^{N}_{+}$, where $N$ is the number of possible assignments across all decision variables. Each number in $\mathcal{H}$ represents a (unnormalized) probability of a particular assignment. To obtain a solution $\mathbf{a}$ from $\mathcal{H}$, one can sample a sequence of assignments from $\mathcal{H}$ while dynamically masking infeasible assignments to meet problem-specific constraints. It can also guide a search process, e.g., Ant Colony Optimization, or be incorporated into hybrid frameworks. Here, the heuristic helps identify promising transitions and improve the efficiency of finding an optimal or near-optimal solution.
+
+
+
+
+
+
+
+## Improvement policies
+
+A policy can be used for improving an initial solution $\mathbf{a}^{0}=(a_{0}^{0},\ldots, a_{T-1}^{0})$ into another one potentially with higher quality, which can be formulated as follows:
+
+$$
+\mathbf{a}^k \sim g(\mathbf{a}^{0}, \mathbf{h}), \quad\pi(\mathbf{a}^K|\mathbf{a}^0,\mathbf{x}) \triangleq \prod_{k=1}^{K-1} g(\mathbf{a}^k | \mathbf{a}^{k-1}, ... ,\mathbf{a}^0, \mathbf{h}),
+$$
+
+where $\mathbf{a}^{k}$ is the $k$-th updated solution and $K$ is the budget for number of improvements. This process allows continuous refinement for a long time to enhance the solution quality.
+
+
+
+
+## Implementation
+
+Policies in our library are subclasses of PyTorch's [`nn.Module`](https://pytorch.org/docs/stable/generated/torch.nn.Module.html) and contain the encoding-decoding logic and neural network parameters $\theta$. Different policies in the RL4CO "zoo" can inherit from metaclasses like `ConstructivePolicy` or `ImprovementPolicy`. We modularize components to process raw features into the embedding space via a parametrized function $\phi_\omega$, called *feature embeddings*.
+
+1. *Node Embeddings $\phi_n$*: transform $m_n$ node features of instances $\mathbf{x}$ from the feature space to the embedding space $h$, i.e., $[B, N, m_n] \rightarrow [B, N, h]$.
+2. *Edge Embeddings $\phi_e$*: transform $m_e$ edge features of instances $\mathbf{x}$ from the feature space to the embedding space $h$, i.e., $[B, E, m_e] \rightarrow [B, E, h]$, where $E$ is the number of edges.
+3. *Context Embeddings $\phi_c$*: capture contextual information by transforming $m_c$ context features from the current decoding step $s_t$ from the feature space to the embedding space $h$, i.e., $[B, m_c] \rightarrow [B, h]$, for nodes or edges.
+
+
+
+
+
+
+ Embeddings can be automatically selected by our library at runtime by simply passing the `env_name` to the policy. Additionally, we allow for granular control of any higher-level policy component independently, such as encoders and decoders.
\ No newline at end of file
diff --git a/docs/content/intro/rl.md b/docs/content/intro/rl.md
new file mode 100644
index 00000000..1e4510cf
--- /dev/null
+++ b/docs/content/intro/rl.md
@@ -0,0 +1,43 @@
+# RL Algorithms
+
+
+## Definitions
+
+The RL objective is to learn a policy $\pi$ that maximizes the expected cumulative reward (or equivalently minimizes the cost) over the distribution of problem instances:
+
+$$
+\theta^{*} = \underset{\theta}{\text{argmax}} \, \mathbb{E}_{\mathbf{x} \sim P(\mathbf{x})} \left[ \mathbb{E}_{\pi(\mathbf{a}|\mathbf{x})} \left[ \sum_{t=0}^{T-1} \gamma^t \mathcal{R}(s_t, a_t) \right] \right],
+$$
+
+where $\theta$ is the set of parameters of $\pi$ and $P(\mathbf{x})$ is the distribution of problem instances.
+
+This equation can be solved using algorithms such as variations of REINFORCE, Advantage Actor-Critic (A2C) methods, or Proximal Policy Optimization (PPO).
+
+These algorithms are employed to train the policy network $\pi$, by transforming the maximization problem into a minimization problem involving a loss function, which is then optimized using gradient descent algorithms. For instance, the REINFORCE loss function gradient is given by:
+
+$$
+\nabla_{\theta} \mathcal{L}_a(\theta|\mathbf{x}) = \mathbb{E}_{\pi(\mathbf{a}|\mathbf{x})} \left[(R(\mathbf{a}, \mathbf{x}) - b(\mathbf{x})) \nabla_{\theta}\log \pi(\mathbf{a}|\mathbf{x})\right],
+$$
+
+where $b(\cdot)$ is a baseline function used to stabilize training and reduce gradient variance.
+
+We also distinguish between two types of RL (pre)training:
+
+1. *Inductive RL*: The focus is on learning patterns from the training dataset to generalize to new instances, thus amortizing the inference procedure.
+2. *Transductive RL* (or test-time optimization): Optimizes parameters during testing on target instances.
+
+Typically, a policy $\pi$ is trained using inductive RL, followed by transductive RL for test-time optimization.
+
+### Implementation
+
+RL algorithms in our library define the process that takes the `Environment` with its problem instances and the `Policy` to optimize its parameters $\theta$. The parent class of algorithms is the `RL4COLitModule`, inheriting from PyTorch Lightning's [`pl.LightningModule`](https://lightning.ai/docs/pytorch/stable/common/lightning_module.html). This allows for granular support of various methods including the `[train, val, test]_step`, automatic logging with several logging services such as Wandb via `log_metrics`, automatic optimizer configuration via `configure_optimizers` and several useful callbacks for RL methods such as `on_train_epoch_end`.
+
+RL algorithms are additionally attached to an `RL4COTrainer`, a wrapper we made with additional optimizations around `pl.Trainer`. This module seamlessly supports features of modern training pipelines, including:
+
+- Logging
+- Checkpoint management
+- Mixed-precision training
+- Various hardware acceleration supports (e.g., CPU, GPU, TPU, and Apple Silicon)
+- Multi-device hardware accelerator in distributed settings
+
+For instance, using mixed-precision training significantly decreases training time without sacrificing much convergence and enables us to leverage recent routines, e.g., FlashAttention.
\ No newline at end of file
diff --git a/docs/_content/start/hydra.md b/docs/content/start/hydra.md
similarity index 90%
rename from docs/_content/start/hydra.md
rename to docs/content/start/hydra.md
index 9126fae2..c34d570c 100644
--- a/docs/_content/start/hydra.md
+++ b/docs/content/start/hydra.md
@@ -9,8 +9,8 @@ Train model with default configuration (AM on TSP environment):
python run.py
```
-> [!TIP]
-> You may check out [this notebook](examples/advanced/1-hydra-config.ipynb) to get started with Hydra!
+!!! tip
+ You may check out [this notebook](examples/advanced/1-hydra-config.ipynb) to get started with Hydra!
### Change experiment
diff --git a/docs/_content/start/installation.md b/docs/content/start/installation.md
similarity index 100%
rename from docs/_content/start/installation.md
rename to docs/content/start/installation.md
diff --git a/docs/hooks.py b/docs/hooks.py
new file mode 100644
index 00000000..24d6640b
--- /dev/null
+++ b/docs/hooks.py
@@ -0,0 +1,52 @@
+import logging
+import os
+import shutil
+from textwrap import dedent
+
+log = logging.getLogger("mkdocs")
+
+# Global variable to store the paths of created __init__.py files
+created_files = []
+
+
+def on_startup(*args, **kwargs):
+ log.info("Creating missing __init__.py files in rl4co package")
+ for subdir, dirs, files in os.walk("rl4co"):
+ if "__init__.py" not in files:
+ init_file_path = os.path.join(subdir, "__init__.py")
+ with open(init_file_path, "w"):
+ pass # empty file
+ created_files.append(init_file_path)
+ log.info(f"{len(created_files)} __init__.py files created")
+
+ # huge trick: we save a backup of README.md and append CSS content to hide some elements
+ log.info("Saving backup of README.md and appending CSS content")
+ shutil.copyfile("README.md", "README_backup.md")
+ # warning: don't touch any of the following. you have been warned :)
+ def append_tricks_to_readme(file_path):
+ # read the tricks from docs/overrides/fancylogo.txt
+ # and put them at the beginning of the file
+ with open("docs/overrides/fancylogo.txt", 'r') as fancylogo:
+ tricks = fancylogo.read()
+ if not os.path.exists(file_path):
+ print(f"Error: The file {file_path} does not exist.")
+ return
+ with open(file_path, 'r') as original:
+ data = original.read()
+ # remove first 33 lines. yeah, it's a hack to remove unneded stuff lol
+ data = '\n'.join(data.split('\n')[33:])
+ with open(file_path, 'w') as modified:
+ modified.write(tricks + data)
+ print(f"CSS content has been appended to {file_path}")
+
+ append_tricks_to_readme("README.md")
+
+
+def on_shutdown(*args, **kwargs):
+ log.info(f"Removing {len(created_files)} created __init__.py files")
+ for file_path in created_files:
+ if os.path.exists(file_path):
+ os.remove(file_path)
+
+ log.info("Replace README.md with README_backup.md")
+ shutil.move("README_backup.md", "README.md")
\ No newline at end of file
diff --git a/docs/index.md b/docs/index.md
deleted file mode 100644
index 87aae5fd..00000000
--- a/docs/index.md
+++ /dev/null
@@ -1,121 +0,0 @@
-# RL4CO
-
-
-
-
-
-
-
-
-
-An extensive Reinforcement Learning (RL) for Combinatorial Optimization (CO) benchmark. Our goal is to provide a unified framework for RL-based CO algorithms, and to facilitate reproducible research in this field, decoupling the science from the engineering.
-
-
-
-
- [](https://github.com/psf/black)
-
-[](https://opensource.org/licenses/MIT) [](https://pypi.org/project/rl4co)
-[](https://github.com/ai4co/rl4co/actions/workflows/tests.yml)
-
-
-
-
-
-
-RL4CO is built upon:
-- [TorchRL](https://github.com/pytorch/rl): official PyTorch framework for RL algorithms and vectorized environments on GPUs
-- [TensorDict](https://github.com/pytorch-labs/tensordict): a library to easily handle heterogeneous data such as states, actions and rewards
-- [PyTorch Lightning](https://github.com/Lightning-AI/lightning): a lightweight PyTorch wrapper for high-performance AI research
-- [Hydra](https://github.com/facebookresearch/hydra): a framework for elegantly configuring complex applications
-
-
-
-
-```{eval-rst}
-.. toctree::
- :maxdepth: 3
- :caption: Getting started
-
- _content/start/installation
- _collections/1-quickstart
- _content/start/hydra
- _content/api/models/common/__init__
-
-
-.. toctree::
- :maxdepth: 3
- :caption: Tutorials
-
- _collections/2-full-training
- _collections/3-creating-new-env-model
- _collections/modeling/1-decoding-strategies
- _collections/modeling/2-transductive-methods
- _collections/modeling/3-change-encoder
- _collections/advanced/1-hydra-config
-
-.. toctree::
- :maxdepth: 3
- :caption: RL Algorithms
-
- _content/api/algos/base
- _content/api/algos/reinforce
- _content/api/algos/a2c
- _content/api/algos/ppo
-
-
-.. toctree::
- :maxdepth: 3
- :caption: Environments
-
- _content/api/envs/base
- _content/api/envs/eda
- _content/api/envs/routing
- _content/api/envs/scheduling
-
-.. toctree::
- :maxdepth: 3
- :caption: Networks
-
- _content/api/models/nn
- _content/api/models/env_embeddings
-
-.. toctree::
- :maxdepth: 3
- :caption: Base NCO Methods
-
- _content/api/models/common/constructive
- _content/api/models/common/improvement
- _content/api/models/common/transductive
-
-
-.. toctree::
- :maxdepth: 3
- :caption: Model Zoo
-
- _content/api/models/common/zoo/__init__
- _content/api/models/common/zoo/constructive
- _content/api/models/common/zoo/improvement
- _content/api/models/common/zoo/transductive
-
-.. toctree::
- :maxdepth: 3
- :caption: Additional API
-
- _content/api/tasks
- _content/api/decoding
- _content/api/data
-
-.. toctree::
- :maxdepth: 3
- :caption: Other
-
- _content/general/contribute
- _content/general/faq
-```
-
-
-## Contributors
-
-
-
\ No newline at end of file
diff --git a/docs/js/autolink.js b/docs/js/autolink.js
new file mode 100644
index 00000000..30a295a5
--- /dev/null
+++ b/docs/js/autolink.js
@@ -0,0 +1,34 @@
+const convertLinks = ( input ) => {
+
+ let text = input;
+ const linksFound = text.match( /(?:www|https?)[^\s]+/g );
+ const aLink = [];
+
+ if ( linksFound != null ) {
+
+ for ( let i=0; i' )
+ }
+ else if ( linkText.match( /vimeo/ ) ) {
+ let vimeoID = replace.split( '/' ).slice(-1)[0];
+ aLink.push( '' )
+ }
+ else {
+ aLink.push( '' + linkText + '' );
+ }
+ text = text.split( linksFound[i] ).map(item => { return aLink[i].includes('iframe') ? item.trim() : item } ).join( aLink[i] );
+ }
+ return text;
+
+ }
+ else {
+ return input;
+ }
+ }
\ No newline at end of file
diff --git a/docs/js/katex.js b/docs/js/katex.js
new file mode 100644
index 00000000..841e35ad
--- /dev/null
+++ b/docs/js/katex.js
@@ -0,0 +1,10 @@
+document$.subscribe(({ body }) => {
+ renderMathInElement(body, {
+ delimiters: [
+ { left: "$$", right: "$$", display: true },
+ { left: "$", right: "$", display: false },
+ { left: "\\(", right: "\\)", display: false },
+ { left: "\\[", right: "\\]", display: true }
+ ],
+ })
+})
\ No newline at end of file
diff --git a/docs/js/tsparticles.js b/docs/js/tsparticles.js
new file mode 100644
index 00000000..bf96e09b
--- /dev/null
+++ b/docs/js/tsparticles.js
@@ -0,0 +1,2 @@
+/*! For license information please see tsparticles.bundle.min.js.LICENSE.txt */
+!function(t,e){if("object"==typeof exports&&"object"==typeof module)module.exports=e();else if("function"==typeof define&&define.amd)define([],e);else{var i=e();for(var s in i)("object"==typeof exports?exports:t)[s]=i[s]}}(this,(()=>(()=>{"use strict";var t={d:(e,i)=>{for(var s in i)t.o(i,s)&&!t.o(e,s)&&Object.defineProperty(e,s,{enumerable:!0,get:i[s]})},o:(t,e)=>Object.prototype.hasOwnProperty.call(t,e),r:t=>{"undefined"!=typeof Symbol&&Symbol.toStringTag&&Object.defineProperty(t,Symbol.toStringTag,{value:"Module"}),Object.defineProperty(t,"__esModule",{value:!0})}},e={};t.r(e),t.d(e,{AnimatableColor:()=>Se,AnimationOptions:()=>Me,AnimationValueWithRandom:()=>Ee,Background:()=>le,BackgroundMask:()=>de,BackgroundMaskCover:()=>he,Circle:()=>yi,ClickEvent:()=>pe,Collisions:()=>Fe,CollisionsAbsorb:()=>De,CollisionsOverlap:()=>Te,ColorAnimation:()=>Pe,DivEvent:()=>fe,Events:()=>ge,ExternalInteractorBase:()=>Si,FullScreen:()=>ue,HoverEvent:()=>ye,HslAnimation:()=>Oe,HslColorManager:()=>Pi,Interactivity:()=>we,ManualParticle:()=>xe,Modes:()=>be,Move:()=>Xe,MoveAngle:()=>qe,MoveAttract:()=>He,MoveCenter:()=>Ve,MoveGravity:()=>Ue,MovePath:()=>We,MoveTrail:()=>je,Opacity:()=>Ze,OpacityAnimation:()=>Ye,Options:()=>li,OptionsColor:()=>ce,OutModes:()=>Ge,Parallax:()=>ve,ParticlesBounce:()=>Ae,ParticlesBounceFactor:()=>Le,ParticlesDensity:()=>Qe,ParticlesInteractorBase:()=>Di,ParticlesNumber:()=>Ke,ParticlesNumberLimit:()=>Je,ParticlesOptions:()=>ai,Point:()=>pi,Range:()=>fi,RangedAnimationOptions:()=>Ce,RangedAnimationValueWithRandom:()=>Ie,Rectangle:()=>vi,ResizeEvent:()=>me,Responsive:()=>_e,RgbColorManager:()=>Oi,Shadow:()=>ti,Shape:()=>ei,Size:()=>si,SizeAnimation:()=>ii,Spin:()=>Ne,Stroke:()=>oi,Theme:()=>ze,ThemeDefault:()=>ke,ValueWithRandom:()=>Re,Vector:()=>y,Vector3d:()=>v,ZIndex:()=>ni,addColorManager:()=>Pt,addEasing:()=>b,alterHsl:()=>se,areBoundsInside:()=>et,arrayRandomIndex:()=>J,calcExactPositionOrRandomFromSize:()=>B,calcExactPositionOrRandomFromSizeRanged:()=>q,calcPositionFromSize:()=>L,calcPositionOrRandomFromSize:()=>A,calcPositionOrRandomFromSizeRanged:()=>F,calculateBounds:()=>it,circleBounce:()=>lt,circleBounceDataFromParticle:()=>ct,clamp:()=>k,clear:()=>Zt,collisionVelocity:()=>I,colorMix:()=>Vt,colorToHsl:()=>Tt,colorToRgb:()=>Dt,deepExtend:()=>st,divMode:()=>rt,divModeExecute:()=>nt,drawEffect:()=>Jt,drawLine:()=>Nt,drawParticle:()=>Qt,drawParticlePlugin:()=>ie,drawPlugin:()=>ee,drawShape:()=>Kt,drawShapeAfterDraw:()=>te,errorPrefix:()=>f,executeOnSingleOrMultiple:()=>dt,findItemFromSingleOrMultiple:()=>pt,generatedAttribute:()=>i,getDistance:()=>T,getDistances:()=>D,getEasing:()=>w,getHslAnimationFromHsl:()=>jt,getHslFromAnimation:()=>$t,getLinkColor:()=>Ut,getLinkRandomColor:()=>Wt,getLogger:()=>W,getParticleBaseVelocity:()=>E,getParticleDirectionAngle:()=>R,getPosition:()=>yt,getRandom:()=>_,getRandomRgbColor:()=>Bt,getRangeMax:()=>O,getRangeMin:()=>P,getRangeValue:()=>C,getSize:()=>mt,getStyleFromHsl:()=>Ht,getStyleFromRgb:()=>qt,hasMatchMedia:()=>G,hslToRgb:()=>At,hslaToRgba:()=>Ft,initParticleNumericAnimationValue:()=>ft,isArray:()=>kt,isBoolean:()=>gt,isDivModeEnabled:()=>ot,isFunction:()=>xt,isInArray:()=>Z,isNumber:()=>wt,isObject:()=>_t,isPointInside:()=>tt,isSsr:()=>j,isString:()=>bt,itemFromArray:()=>K,itemFromSingleOrMultiple:()=>ut,loadFont:()=>Q,loadFull:()=>gn,loadOptions:()=>ri,loadParticlesOptions:()=>ci,loadSlim:()=>an,mix:()=>z,mouseDownEvent:()=>s,mouseLeaveEvent:()=>n,mouseMoveEvent:()=>r,mouseOutEvent:()=>a,mouseUpEvent:()=>o,paintBase:()=>Xt,paintImage:()=>Yt,parseAlpha:()=>H,randomInRange:()=>M,rangeColorToHsl:()=>Rt,rangeColorToRgb:()=>St,rectBounce:()=>ht,resizeEvent:()=>u,rgbToHsl:()=>Et,safeIntersectionObserver:()=>X,safeMatchMedia:()=>N,safeMutationObserver:()=>Y,setLogger:()=>U,setRandom:()=>x,setRangeValue:()=>S,singleDivModeExecute:()=>at,stringToAlpha:()=>It,stringToRgb:()=>Lt,touchCancelEvent:()=>d,touchEndEvent:()=>l,touchMoveEvent:()=>h,touchStartEvent:()=>c,tsParticles:()=>Ti,visibilityChangeEvent:()=>p});const i="generated",s="pointerdown",o="pointerup",n="pointerleave",a="pointerout",r="pointermove",c="touchstart",l="touchend",h="touchmove",d="touchcancel",u="resize",p="visibilitychange",f="tsParticles - Error";class v{constructor(t,e,i){if(this._updateFromAngle=(t,e)=>{this.x=Math.cos(t)*e,this.y=Math.sin(t)*e},!wt(t)&&t){this.x=t.x,this.y=t.y;const e=t;this.z=e.z?e.z:0}else{if(void 0===t||void 0===e)throw new Error(`${f} Vector3d not initialized correctly`);this.x=t,this.y=e,this.z=i??0}}static get origin(){return v.create(0,0,0)}get angle(){return Math.atan2(this.y,this.x)}set angle(t){this._updateFromAngle(t,this.length)}get length(){return Math.sqrt(this.getLengthSq())}set length(t){this._updateFromAngle(this.angle,t)}static clone(t){return v.create(t.x,t.y,t.z)}static create(t,e,i){return new v(t,e,i)}add(t){return v.create(this.x+t.x,this.y+t.y,this.z+t.z)}addTo(t){this.x+=t.x,this.y+=t.y,this.z+=t.z}copy(){return v.clone(this)}distanceTo(t){return this.sub(t).length}distanceToSq(t){return this.sub(t).getLengthSq()}div(t){return v.create(this.x/t,this.y/t,this.z/t)}divTo(t){this.x/=t,this.y/=t,this.z/=t}getLengthSq(){return this.x**2+this.y**2}mult(t){return v.create(this.x*t,this.y*t,this.z*t)}multTo(t){this.x*=t,this.y*=t,this.z*=t}normalize(){const t=this.length;0!=t&&this.multTo(1/t)}rotate(t){return v.create(this.x*Math.cos(t)-this.y*Math.sin(t),this.x*Math.sin(t)+this.y*Math.cos(t),0)}setTo(t){this.x=t.x,this.y=t.y;const e=t;this.z=e.z?e.z:0}sub(t){return v.create(this.x-t.x,this.y-t.y,this.z-t.z)}subFrom(t){this.x-=t.x,this.y-=t.y,this.z-=t.z}}class y extends v{constructor(t,e){super(t,e,0)}static get origin(){return y.create(0,0)}static clone(t){return y.create(t.x,t.y)}static create(t,e){return new y(t,e)}}let m=Math.random;const g=new Map;function b(t,e){g.get(t)||g.set(t,e)}function w(t){return g.get(t)||(t=>t)}function x(t=Math.random){m=t}function _(){return k(m(),0,1-1e-16)}function k(t,e,i){return Math.min(Math.max(t,e),i)}function z(t,e,i,s){return Math.floor((t*i+e*s)/(i+s))}function M(t){const e=O(t);let i=P(t);return e===i&&(i=0),_()*(e-i)+i}function C(t){return wt(t)?t:M(t)}function P(t){return wt(t)?t:t.min}function O(t){return wt(t)?t:t.max}function S(t,e){if(t===e||void 0===e&&wt(t))return t;const i=P(t),s=O(t);return void 0!==e?{min:Math.min(i,e),max:Math.max(s,e)}:S(i,s)}function D(t,e){const i=t.x-e.x,s=t.y-e.y;return{dx:i,dy:s,distance:Math.sqrt(i**2+s**2)}}function T(t,e){return D(t,e).distance}function R(t,e,i){if(wt(t))return t*Math.PI/180;switch(t){case"top":return.5*-Math.PI;case"top-right":return.25*-Math.PI;case"right":return 0;case"bottom-right":return.25*Math.PI;case"bottom":return.5*Math.PI;case"bottom-left":return.75*Math.PI;case"left":return Math.PI;case"top-left":return.75*-Math.PI;case"inside":return Math.atan2(i.y-e.y,i.x-e.x);case"outside":return Math.atan2(e.y-i.y,e.x-i.x);default:return _()*Math.PI*2}}function E(t){const e=y.origin;return e.length=1,e.angle=t,e}function I(t,e,i,s){return y.create(t.x*(i-s)/(i+s)+2*e.x*s/(i+s),t.y)}function L(t){return t.position&&void 0!==t.position.x&&void 0!==t.position.y?{x:t.position.x*t.size.width/100,y:t.position.y*t.size.height/100}:void 0}function A(t){return{x:(t.position?.x??100*_())*t.size.width/100,y:(t.position?.y??100*_())*t.size.height/100}}function F(t){const e={x:void 0!==t.position?.x?C(t.position.x):void 0,y:void 0!==t.position?.y?C(t.position.y):void 0};return A({size:t.size,position:e})}function B(t){return{x:t.position?.x??_()*t.size.width,y:t.position?.y??_()*t.size.height}}function q(t){const e={x:void 0!==t.position?.x?C(t.position.x):void 0,y:void 0!==t.position?.y?C(t.position.y):void 0};return B({size:t.size,position:e})}function H(t){return t?t.endsWith("%")?parseFloat(t)/100:parseFloat(t):1}const V={debug:console.debug,error:console.error,info:console.info,log:console.log,verbose:console.log,warning:console.warn};function U(t){V.debug=t.debug||V.debug,V.error=t.error||V.error,V.info=t.info||V.info,V.log=t.log||V.log,V.verbose=t.verbose||V.verbose,V.warning=t.warning||V.warning}function W(){return V}function $(t){const e={bounced:!1},{pSide:i,pOtherSide:s,rectSide:o,rectOtherSide:n,velocity:a,factor:r}=t;return s.minn.max||s.maxn.max||(i.max>=o.min&&i.max<=.5*(o.max+o.min)&&a>0||i.min<=o.max&&i.min>.5*(o.max+o.min)&&a<0)&&(e.velocity=a*-r,e.bounced=!0),e}function j(){return"undefined"==typeof window||!window||void 0===window.document||!window.document}function G(){return!j()&&"undefined"!=typeof matchMedia}function N(t){if(G())return matchMedia(t)}function X(t){if(!j()&&"undefined"!=typeof IntersectionObserver)return new IntersectionObserver(t)}function Y(t){if(!j()&&"undefined"!=typeof MutationObserver)return new MutationObserver(t)}function Z(t,e){return t===e||kt(e)&&e.indexOf(t)>-1}async function Q(t,e){try{await document.fonts.load(`${e??"400"} 36px '${t??"Verdana"}'`)}catch{}}function J(t){return Math.floor(_()*t.length)}function K(t,e,i=!0){return t[void 0!==e&&i?e%t.length:J(t)]}function tt(t,e,i,s,o){return et(it(t,s??0),e,i,o)}function et(t,e,i,s){let o=!0;return s&&"bottom"!==s||(o=t.topi.x),!o||s&&"right"!==s||(o=t.lefti.y),o}function it(t,e){return{bottom:t.y+e,left:t.x-e,right:t.x+e,top:t.y-e}}function st(t,...e){for(const i of e){if(null==i)continue;if(!_t(i)){t=i;continue}const e=Array.isArray(i);!e||!_t(t)&&t&&Array.isArray(t)?e||!_t(t)&&t&&!Array.isArray(t)||(t={}):t=[];for(const e in i){if("__proto__"===e)continue;const s=i[e],o=t;o[e]=_t(s)&&Array.isArray(s)?s.map((t=>st(o[e],t))):st(o[e],s)}}return t}function ot(t,e){return!!pt(e,(e=>e.enable&&Z(t,e.mode)))}function nt(t,e,i){dt(e,(e=>{const s=e.mode;e.enable&&Z(t,s)&&at(e,i)}))}function at(t,e){dt(t.selectors,(i=>{e(i,t)}))}function rt(t,e){if(e&&t)return pt(t,(t=>function(t,e){const i=dt(e,(e=>t.matches(e)));return kt(i)?i.some((t=>t)):i}(e,t.selectors)))}function ct(t){return{position:t.getPosition(),radius:t.getRadius(),mass:t.getMass(),velocity:t.velocity,factor:y.create(C(t.options.bounce.horizontal.value),C(t.options.bounce.vertical.value))}}function lt(t,e){const{x:i,y:s}=t.velocity.sub(e.velocity),[o,n]=[t.position,e.position],{dx:a,dy:r}=D(n,o);if(i*a+s*r<0)return;const c=-Math.atan2(r,a),l=t.mass,h=e.mass,d=t.velocity.rotate(c),u=e.velocity.rotate(c),p=I(d,u,l,h),f=I(u,d,l,h),v=p.rotate(-c),y=f.rotate(-c);t.velocity.x=v.x*t.factor.x,t.velocity.y=v.y*t.factor.y,e.velocity.x=y.x*e.factor.x,e.velocity.y=y.y*e.factor.y}function ht(t,e){const i=it(t.getPosition(),t.getRadius()),s=t.options.bounce,o=$({pSide:{min:i.left,max:i.right},pOtherSide:{min:i.top,max:i.bottom},rectSide:{min:e.left,max:e.right},rectOtherSide:{min:e.top,max:e.bottom},velocity:t.velocity.x,factor:C(s.horizontal.value)});o.bounced&&(void 0!==o.velocity&&(t.velocity.x=o.velocity),void 0!==o.position&&(t.position.x=o.position));const n=$({pSide:{min:i.top,max:i.bottom},pOtherSide:{min:i.left,max:i.right},rectSide:{min:e.top,max:e.bottom},rectOtherSide:{min:e.left,max:e.right},velocity:t.velocity.y,factor:C(s.vertical.value)});n.bounced&&(void 0!==n.velocity&&(t.velocity.y=n.velocity),void 0!==n.position&&(t.position.y=n.position))}function dt(t,e){return kt(t)?t.map(((t,i)=>e(t,i))):e(t,0)}function ut(t,e,i){return kt(t)?K(t,e,i):t}function pt(t,e){return kt(t)?t.find(((t,i)=>e(t,i))):e(t,0)?t:void 0}function ft(t,e){const i=t.value,s=t.animation,o={delayTime:1e3*C(s.delay),enable:s.enable,value:C(t.value)*e,max:O(i)*e,min:P(i)*e,loops:0,maxLoops:C(s.count),time:0};if(s.enable){switch(o.decay=1-C(s.decay),s.mode){case"increase":o.status="increasing";break;case"decrease":o.status="decreasing";break;case"random":o.status=_()>=.5?"increasing":"decreasing"}const t="auto"===s.mode;switch(s.startValue){case"min":o.value=o.min,t&&(o.status="increasing");break;case"max":o.value=o.max,t&&(o.status="decreasing");break;default:o.value=M(o),t&&(o.status=_()>=.5?"increasing":"decreasing")}}return o.initialValue=o.value,o}function vt(t,e){if(!("percent"===t.mode)){const{mode:e,...i}=t;return i}return"x"in t?{x:t.x/100*e.width,y:t.y/100*e.height}:{width:t.width/100*e.width,height:t.height/100*e.height}}function yt(t,e){return vt(t,e)}function mt(t,e){return vt(t,e)}function gt(t){return"boolean"==typeof t}function bt(t){return"string"==typeof t}function wt(t){return"number"==typeof t}function xt(t){return"function"==typeof t}function _t(t){return"object"==typeof t&&null!==t}function kt(t){return Array.isArray(t)}const zt="random",Mt="mid",Ct=new Map;function Pt(t){Ct.set(t.key,t)}function Ot(t){for(const[,e]of Ct)if(t.startsWith(e.stringPrefix))return e.parseString(t);const e=t.replace(/^#?([a-f\d])([a-f\d])([a-f\d])([a-f\d])?$/i,((t,e,i,s,o)=>e+e+i+i+s+s+(void 0!==o?o+o:""))),i=/^#?([a-f\d]{2})([a-f\d]{2})([a-f\d]{2})([a-f\d]{2})?$/i.exec(e);return i?{a:void 0!==i[4]?parseInt(i[4],16)/255:1,b:parseInt(i[3],16),g:parseInt(i[2],16),r:parseInt(i[1],16)}:void 0}function St(t,e,i=!0){if(!t)return;const s=bt(t)?{value:t}:t;if(bt(s.value))return Dt(s.value,e,i);if(kt(s.value))return St({value:K(s.value,e,i)});for(const[,t]of Ct){const e=t.handleRangeColor(s);if(e)return e}}function Dt(t,e,i=!0){if(!t)return;const s=bt(t)?{value:t}:t;if(bt(s.value))return s.value===zt?Bt():Lt(s.value);if(kt(s.value))return Dt({value:K(s.value,e,i)});for(const[,t]of Ct){const e=t.handleColor(s);if(e)return e}}function Tt(t,e,i=!0){const s=Dt(t,e,i);return s?Et(s):void 0}function Rt(t,e,i=!0){const s=St(t,e,i);return s?Et(s):void 0}function Et(t){const e=t.r/255,i=t.g/255,s=t.b/255,o=Math.max(e,i,s),n=Math.min(e,i,s),a={h:0,l:.5*(o+n),s:0};return o!==n&&(a.s=a.l<.5?(o-n)/(o+n):(o-n)/(2-o-n),a.h=e===o?(i-s)/(o-n):a.h=i===o?2+(s-e)/(o-n):4+(e-i)/(o-n)),a.l*=100,a.s*=100,a.h*=60,a.h<0&&(a.h+=360),a.h>=360&&(a.h-=360),a}function It(t){return Ot(t)?.a}function Lt(t){return Ot(t)}function At(t){const e=(t.h%360+360)%360,i=Math.max(0,Math.min(100,t.s)),s=e/360,o=i/100,n=Math.max(0,Math.min(100,t.l))/100;if(0===i){const t=Math.round(255*n);return{r:t,g:t,b:t}}const a=(t,e,i)=>(i<0&&(i+=1),i>1&&(i-=1),6*i<1?t+6*(e-t)*i:2*i<1?e:3*i<2?t+(e-t)*(2/3-i)*6:t),r=n<.5?n*(1+o):n+o-n*o,c=2*n-r,l=Math.min(255,255*a(c,r,s+1/3)),h=Math.min(255,255*a(c,r,s)),d=Math.min(255,255*a(c,r,s-1/3));return{r:Math.round(l),g:Math.round(h),b:Math.round(d)}}function Ft(t){const e=At(t);return{a:t.a,b:e.b,g:e.g,r:e.r}}function Bt(t){const e=t??0;return{b:Math.floor(M(S(e,256))),g:Math.floor(M(S(e,256))),r:Math.floor(M(S(e,256)))}}function qt(t,e){return`rgba(${t.r}, ${t.g}, ${t.b}, ${e??1})`}function Ht(t,e){return`hsla(${t.h}, ${t.s}%, ${t.l}%, ${e??1})`}function Vt(t,e,i,s){let o=t,n=e;return void 0===o.r&&(o=At(t)),void 0===n.r&&(n=At(e)),{b:z(o.b,n.b,i,s),g:z(o.g,n.g,i,s),r:z(o.r,n.r,i,s)}}function Ut(t,e,i){if(i===zt)return Bt();if(i!==Mt)return i;{const i=t.getFillColor()??t.getStrokeColor(),s=e?.getFillColor()??e?.getStrokeColor();if(i&&s&&e)return Vt(i,s,t.getRadius(),e.getRadius());{const t=i??s;if(t)return At(t)}}}function Wt(t,e,i){const s=bt(t)?t:t.value;return s===zt?i?St({value:s}):e?zt:Mt:s===Mt?Mt:St({value:s})}function $t(t){return void 0!==t?{h:t.h.value,s:t.s.value,l:t.l.value}:void 0}function jt(t,e,i){const s={h:{enable:!1,value:t.h},s:{enable:!1,value:t.s},l:{enable:!1,value:t.l}};return e&&(Gt(s.h,e.h,i),Gt(s.s,e.s,i),Gt(s.l,e.l,i)),s}function Gt(t,e,i){t.enable=e.enable,t.enable?(t.velocity=C(e.speed)/100*i,t.decay=1-C(e.decay),t.status="increasing",t.loops=0,t.maxLoops=C(e.count),t.time=0,t.delayTime=1e3*C(e.delay),e.sync||(t.velocity*=_(),t.value*=_()),t.initialValue=t.value):t.velocity=0}function Nt(t,e,i){t.beginPath(),t.moveTo(e.x,e.y),t.lineTo(i.x,i.y),t.closePath()}function Xt(t,e,i){t.fillStyle=i??"rgba(0,0,0,0)",t.fillRect(0,0,e.width,e.height)}function Yt(t,e,i,s){i&&(t.globalAlpha=s,t.drawImage(i,0,0,e.width,e.height),t.globalAlpha=1)}function Zt(t,e){t.clearRect(0,0,e.width,e.height)}function Qt(t){const{container:e,context:i,particle:s,delta:o,colorStyles:n,backgroundMask:a,composite:r,radius:c,opacity:l,shadow:h,transform:d}=t,u=s.getPosition(),p=s.rotation+(s.pathRotation?s.velocity.angle:0),f=Math.sin(p),v=Math.cos(p),y={a:v*(d.a??1),b:f*(d.b??1),c:-f*(d.c??1),d:v*(d.d??1)};i.setTransform(y.a,y.b,y.c,y.d,u.x,u.y),a&&(i.globalCompositeOperation=r);const m=s.shadowColor;h.enable&&m&&(i.shadowBlur=h.blur,i.shadowColor=qt(m),i.shadowOffsetX=h.offset.x,i.shadowOffsetY=h.offset.y),n.fill&&(i.fillStyle=n.fill);const g=s.strokeWidth??0;i.lineWidth=g,n.stroke&&(i.strokeStyle=n.stroke);const b={container:e,context:i,particle:s,radius:c,opacity:l,delta:o,transformData:y};i.beginPath(),Kt(b),s.shapeClose&&i.closePath(),g>0&&i.stroke(),s.shapeFill&&i.fill(),te(b),Jt(b),i.globalCompositeOperation="source-over",i.setTransform(1,0,0,1,0,0)}function Jt(t){const{container:e,context:i,particle:s,radius:o,opacity:n,delta:a,transformData:r}=t;if(!s.effect)return;const c=e.effectDrawers.get(s.effect);c&&c.draw({context:i,particle:s,radius:o,opacity:n,delta:a,pixelRatio:e.retina.pixelRatio,transformData:{...r}})}function Kt(t){const{container:e,context:i,particle:s,radius:o,opacity:n,delta:a,transformData:r}=t;if(!s.shape)return;const c=e.shapeDrawers.get(s.shape);c&&c.draw({context:i,particle:s,radius:o,opacity:n,delta:a,pixelRatio:e.retina.pixelRatio,transformData:{...r}})}function te(t){const{container:e,context:i,particle:s,radius:o,opacity:n,delta:a,transformData:r}=t;if(!s.shape)return;const c=e.shapeDrawers.get(s.shape);c&&c.afterDraw&&c.afterDraw({context:i,particle:s,radius:o,opacity:n,delta:a,pixelRatio:e.retina.pixelRatio,transformData:{...r}})}function ee(t,e,i){e.draw&&e.draw(t,i)}function ie(t,e,i,s){e.drawParticle&&e.drawParticle(t,i,s)}function se(t,e,i){return{h:t.h,s:t.s,l:t.l+("darken"===e?-1:1)*i}}function oe(t,e,i){const s=e[i];void 0!==s&&(t[i]=(t[i]??1)*s)}class ne{constructor(t){this.container=t,this._applyPostDrawUpdaters=t=>{for(const e of this._postDrawUpdaters)e.afterDraw&&e.afterDraw(t)},this._applyPreDrawUpdaters=(t,e,i,s,o,n)=>{for(const a of this._preDrawUpdaters){if(a.getColorStyles){const{fill:n,stroke:r}=a.getColorStyles(e,t,i,s);n&&(o.fill=n),r&&(o.stroke=r)}if(a.getTransformValues){const t=a.getTransformValues(e);for(const e in t)oe(n,t,e)}a.beforeDraw&&a.beforeDraw(e)}},this._applyResizePlugins=()=>{for(const t of this._resizePlugins)t.resize&&t.resize()},this._getPluginParticleColors=t=>{let e,i;for(const s of this._colorPlugins)if(!e&&s.particleFillColor&&(e=Rt(s.particleFillColor(t))),!i&&s.particleStrokeColor&&(i=Rt(s.particleStrokeColor(t))),e&&i)break;return[e,i]},this._initCover=()=>{const t=this.container.actualOptions.backgroundMask.cover,e=St(t.color);if(e){const i={...e,a:t.opacity};this._coverColorStyle=qt(i,i.a)}},this._initStyle=()=>{const t=this.element,e=this.container.actualOptions;if(t){this._fullScreen?(this._originalStyle=st({},t.style),this._setFullScreenStyle()):this._resetOriginalStyle();for(const i in e.style){if(!i||!e.style)continue;const s=e.style[i];s&&t.style.setProperty(i,s,"important")}}},this._initTrail=async()=>{const t=this.container.actualOptions,e=t.particles.move.trail,i=e.fill;if(e.enable)if(i.color){const e=St(i.color);if(!e)return;const s=t.particles.move.trail;this._trailFill={color:{...e},opacity:1/s.length}}else await new Promise(((t,s)=>{if(!i.image)return;const o=document.createElement("img");o.addEventListener("load",(()=>{this._trailFill={image:o,opacity:1/e.length},t()})),o.addEventListener("error",(t=>{s(t.error)})),o.src=i.image}))},this._paintBase=t=>{this.draw((e=>Xt(e,this.size,t)))},this._paintImage=(t,e)=>{this.draw((i=>Yt(i,this.size,t,e)))},this._repairStyle=()=>{const t=this.element;t&&(this._safeMutationObserver((t=>t.disconnect())),this._initStyle(),this.initBackground(),this._safeMutationObserver((e=>e.observe(t,{attributes:!0}))))},this._resetOriginalStyle=()=>{const t=this.element,e=this._originalStyle;if(!t||!e)return;const i=t.style;i.position=e.position,i.zIndex=e.zIndex,i.top=e.top,i.left=e.left,i.width=e.width,i.height=e.height},this._safeMutationObserver=t=>{this._mutationObserver&&t(this._mutationObserver)},this._setFullScreenStyle=()=>{const t=this.element;if(!t)return;const e="important",i=t.style;i.setProperty("position","fixed",e),i.setProperty("z-index",this.container.actualOptions.fullScreen.zIndex.toString(10),e),i.setProperty("top","0",e),i.setProperty("left","0",e),i.setProperty("width","100%",e),i.setProperty("height","100%",e)},this.size={height:0,width:0},this._context=null,this._generated=!1,this._preDrawUpdaters=[],this._postDrawUpdaters=[],this._resizePlugins=[],this._colorPlugins=[]}get _fullScreen(){return this.container.actualOptions.fullScreen.enable}clear(){const t=this.container.actualOptions,e=t.particles.move.trail,i=this._trailFill;t.backgroundMask.enable?this.paint():e.enable&&e.length>0&&i?i.color?this._paintBase(qt(i.color,i.opacity)):i.image&&this._paintImage(i.image,i.opacity):t.clear&&this.draw((t=>{Zt(t,this.size)}))}destroy(){if(this.stop(),this._generated){const t=this.element;t&&t.remove()}else this._resetOriginalStyle();this._preDrawUpdaters=[],this._postDrawUpdaters=[],this._resizePlugins=[],this._colorPlugins=[]}draw(t){const e=this._context;if(e)return t(e)}drawParticle(t,e){if(t.spawning||t.destroyed)return;const i=t.getRadius();if(i<=0)return;const s=t.getFillColor(),o=t.getStrokeColor()??s;let[n,a]=this._getPluginParticleColors(t);n||(n=s),a||(a=o),(n||a)&&this.draw((s=>{const o=this.container,r=o.actualOptions,c=t.options.zIndex,l=(1-t.zIndexFactor)**c.opacityRate,h=t.bubble.opacity??t.opacity?.value??1,d=h*l,u=(t.strokeOpacity??h)*l,p={},f={fill:n?Ht(n,d):void 0};f.stroke=a?Ht(a,u):f.fill,this._applyPreDrawUpdaters(s,t,i,d,f,p),Qt({container:o,context:s,particle:t,delta:e,colorStyles:f,backgroundMask:r.backgroundMask.enable,composite:r.backgroundMask.composite,radius:i*(1-t.zIndexFactor)**c.sizeRate,opacity:d,shadow:t.options.shadow,transform:p}),this._applyPostDrawUpdaters(t)}))}drawParticlePlugin(t,e,i){this.draw((s=>ie(s,t,e,i)))}drawPlugin(t,e){this.draw((i=>ee(i,t,e)))}async init(){this._safeMutationObserver((t=>t.disconnect())),this._mutationObserver=Y((t=>{for(const e of t)"attributes"===e.type&&"style"===e.attributeName&&this._repairStyle()})),this.resize(),this._initStyle(),this._initCover();try{await this._initTrail()}catch(t){W().error(t)}this.initBackground(),this._safeMutationObserver((t=>{this.element&&t.observe(this.element,{attributes:!0})})),this.initUpdaters(),this.initPlugins(),this.paint()}initBackground(){const t=this.container.actualOptions.background,e=this.element;if(!e)return;const i=e.style;if(i){if(t.color){const e=St(t.color);i.backgroundColor=e?qt(e,t.opacity):""}else i.backgroundColor="";i.backgroundImage=t.image||"",i.backgroundPosition=t.position||"",i.backgroundRepeat=t.repeat||"",i.backgroundSize=t.size||""}}initPlugins(){this._resizePlugins=[];for(const[,t]of this.container.plugins)t.resize&&this._resizePlugins.push(t),(t.particleFillColor||t.particleStrokeColor)&&this._colorPlugins.push(t)}initUpdaters(){this._preDrawUpdaters=[],this._postDrawUpdaters=[];for(const t of this.container.particles.updaters)t.afterDraw&&this._postDrawUpdaters.push(t),(t.getColorStyles||t.getTransformValues||t.beforeDraw)&&this._preDrawUpdaters.push(t)}loadCanvas(t){this._generated&&this.element&&this.element.remove(),this._generated=t.dataset&&i in t.dataset?"true"===t.dataset[i]:this._generated,this.element=t,this.element.ariaHidden="true",this._originalStyle=st({},this.element.style),this.size.height=t.offsetHeight,this.size.width=t.offsetWidth,this._context=this.element.getContext("2d"),this._safeMutationObserver((t=>{this.element&&t.observe(this.element,{attributes:!0})})),this.container.retina.init(),this.initBackground()}paint(){const t=this.container.actualOptions;this.draw((e=>{t.backgroundMask.enable&&t.backgroundMask.cover?(Zt(e,this.size),this._paintBase(this._coverColorStyle)):this._paintBase()}))}resize(){if(!this.element)return!1;const t=this.container,e=t.retina.pixelRatio,i=t.canvas.size,s=this.element.offsetWidth*e,o=this.element.offsetHeight*e;if(o===i.height&&s===i.width&&o===this.element.height&&s===this.element.width)return!1;const n={...i};return this.element.width=i.width=this.element.offsetWidth*e,this.element.height=i.height=this.element.offsetHeight*e,this.container.started&&t.particles.setResizeFactor({width:i.width/n.width,height:i.height/n.height}),!0}stop(){this._safeMutationObserver((t=>t.disconnect())),this._mutationObserver=void 0,this.draw((t=>Zt(t,this.size)))}async windowResize(){if(!this.element||!this.resize())return;const t=this.container,e=t.updateActualOptions();t.particles.setDensity(),this._applyResizePlugins(),e&&await t.refresh()}}function ae(t,e,i,s,o){if(s){let s={passive:!0};gt(o)?s.capture=o:void 0!==o&&(s=o),t.addEventListener(e,i,s)}else{const s=o;t.removeEventListener(e,i,s)}}class re{constructor(t){this.container=t,this._doMouseTouchClick=t=>{const e=this.container,i=e.actualOptions;if(this._canPush){const t=e.interactivity.mouse,s=t.position;if(!s)return;t.clickPosition={...s},t.clickTime=(new Date).getTime();dt(i.interactivity.events.onClick.mode,(t=>this.container.handleClickMode(t)))}"touchend"===t.type&&setTimeout((()=>this._mouseTouchFinish()),500)},this._handleThemeChange=t=>{const e=t,i=this.container,s=i.options,o=s.defaultThemes,n=e.matches?o.dark:o.light,a=s.themes.find((t=>t.name===n));a&&a.default.auto&&i.loadTheme(n)},this._handleVisibilityChange=()=>{const t=this.container,e=t.actualOptions;this._mouseTouchFinish(),e.pauseOnBlur&&(document&&document.hidden?(t.pageHidden=!0,t.pause()):(t.pageHidden=!1,t.getAnimationStatus()?t.play(!0):t.draw(!0)))},this._handleWindowResize=async()=>{this._resizeTimeout&&(clearTimeout(this._resizeTimeout),delete this._resizeTimeout),this._resizeTimeout=setTimeout((async()=>{const t=this.container.canvas;t&&await t.windowResize()}),1e3*this.container.actualOptions.interactivity.events.resize.delay)},this._manageInteractivityListeners=(t,e)=>{const i=this._handlers,n=this.container,a=n.actualOptions,u=n.interactivity.element;if(!u)return;const p=u,f=n.canvas.element;f&&(f.style.pointerEvents=p===f?"initial":"none"),(a.interactivity.events.onHover.enable||a.interactivity.events.onClick.enable)&&(ae(u,r,i.mouseMove,e),ae(u,c,i.touchStart,e),ae(u,h,i.touchMove,e),a.interactivity.events.onClick.enable?(ae(u,l,i.touchEndClick,e),ae(u,o,i.mouseUp,e),ae(u,s,i.mouseDown,e)):ae(u,l,i.touchEnd,e),ae(u,t,i.mouseLeave,e),ae(u,d,i.touchCancel,e))},this._manageListeners=t=>{const e=this._handlers,i=this.container,s=i.actualOptions.interactivity.detectsOn,o=i.canvas.element;let r=n;"window"===s?(i.interactivity.element=window,r=a):i.interactivity.element="parent"===s&&o?o.parentElement??o.parentNode:o,this._manageMediaMatch(t),this._manageResize(t),this._manageInteractivityListeners(r,t),document&&ae(document,p,e.visibilityChange,t,!1)},this._manageMediaMatch=t=>{const e=this._handlers,i=N("(prefers-color-scheme: dark)");i&&(void 0===i.addEventListener?void 0!==i.addListener&&(t?i.addListener(e.oldThemeChange):i.removeListener(e.oldThemeChange)):ae(i,"change",e.themeChange,t))},this._manageResize=t=>{const e=this._handlers,i=this.container;if(!i.actualOptions.interactivity.events.resize)return;if("undefined"==typeof ResizeObserver)return void ae(window,u,e.resize,t);const s=i.canvas.element;this._resizeObserver&&!t?(s&&this._resizeObserver.unobserve(s),this._resizeObserver.disconnect(),delete this._resizeObserver):!this._resizeObserver&&t&&s&&(this._resizeObserver=new ResizeObserver((async t=>{t.find((t=>t.target===s))&&await this._handleWindowResize()})),this._resizeObserver.observe(s))},this._mouseDown=()=>{const{interactivity:t}=this.container;if(!t)return;const{mouse:e}=t;e.clicking=!0,e.downPosition=e.position},this._mouseTouchClick=t=>{const e=this.container,i=e.actualOptions,{mouse:s}=e.interactivity;s.inside=!0;let o=!1;const n=s.position;if(n&&i.interactivity.events.onClick.enable){for(const[,t]of e.plugins)if(t.clickPositionValid&&(o=t.clickPositionValid(n),o))break;o||this._doMouseTouchClick(t),s.clicking=!1}},this._mouseTouchFinish=()=>{const t=this.container.interactivity;if(!t)return;const e=t.mouse;delete e.position,delete e.clickPosition,delete e.downPosition,t.status=n,e.inside=!1,e.clicking=!1},this._mouseTouchMove=t=>{const e=this.container,i=e.actualOptions,s=e.interactivity,o=e.canvas.element;if(!s||!s.element)return;let n;if(s.mouse.inside=!0,t.type.startsWith("pointer")){this._canPush=!0;const e=t;if(s.element===window){if(o){const t=o.getBoundingClientRect();n={x:e.clientX-t.left,y:e.clientY-t.top}}}else if("parent"===i.interactivity.detectsOn){const t=e.target,i=e.currentTarget;if(t&&i&&o){const s=t.getBoundingClientRect(),a=i.getBoundingClientRect(),r=o.getBoundingClientRect();n={x:e.offsetX+2*s.left-(a.left+r.left),y:e.offsetY+2*s.top-(a.top+r.top)}}else n={x:e.offsetX??e.clientX,y:e.offsetY??e.clientY}}else e.target===o&&(n={x:e.offsetX??e.clientX,y:e.offsetY??e.clientY})}else if(this._canPush="touchmove"!==t.type,o){const e=t,i=e.touches[e.touches.length-1],s=o.getBoundingClientRect();n={x:i.clientX-(s.left??0),y:i.clientY-(s.top??0)}}const a=e.retina.pixelRatio;n&&(n.x*=a,n.y*=a),s.mouse.position=n,s.status=r},this._touchEnd=t=>{const e=t,i=Array.from(e.changedTouches);for(const t of i)this._touches.delete(t.identifier);this._mouseTouchFinish()},this._touchEndClick=t=>{const e=t,i=Array.from(e.changedTouches);for(const t of i)this._touches.delete(t.identifier);this._mouseTouchClick(t)},this._touchStart=t=>{const e=t,i=Array.from(e.changedTouches);for(const t of i)this._touches.set(t.identifier,performance.now());this._mouseTouchMove(t)},this._canPush=!0,this._touches=new Map,this._handlers={mouseDown:()=>this._mouseDown(),mouseLeave:()=>this._mouseTouchFinish(),mouseMove:t=>this._mouseTouchMove(t),mouseUp:t=>this._mouseTouchClick(t),touchStart:t=>this._touchStart(t),touchMove:t=>this._mouseTouchMove(t),touchEnd:t=>this._touchEnd(t),touchCancel:t=>this._touchEnd(t),touchEndClick:t=>this._touchEndClick(t),visibilityChange:()=>this._handleVisibilityChange(),themeChange:t=>this._handleThemeChange(t),oldThemeChange:t=>this._handleThemeChange(t),resize:()=>{this._handleWindowResize()}}}addListeners(){this._manageListeners(!0)}removeListeners(){this._manageListeners(!1)}}class ce{constructor(){this.value=""}static create(t,e){const i=new ce;return i.load(t),void 0!==e&&(bt(e)||kt(e)?i.load({value:e}):i.load(e)),i}load(t){void 0!==t?.value&&(this.value=t.value)}}class le{constructor(){this.color=new ce,this.color.value="",this.image="",this.position="",this.repeat="",this.size="",this.opacity=1}load(t){t&&(void 0!==t.color&&(this.color=ce.create(this.color,t.color)),void 0!==t.image&&(this.image=t.image),void 0!==t.position&&(this.position=t.position),void 0!==t.repeat&&(this.repeat=t.repeat),void 0!==t.size&&(this.size=t.size),void 0!==t.opacity&&(this.opacity=t.opacity))}}class he{constructor(){this.color=new ce,this.color.value="#fff",this.opacity=1}load(t){t&&(void 0!==t.color&&(this.color=ce.create(this.color,t.color)),void 0!==t.opacity&&(this.opacity=t.opacity))}}class de{constructor(){this.composite="destination-out",this.cover=new he,this.enable=!1}load(t){if(t){if(void 0!==t.composite&&(this.composite=t.composite),void 0!==t.cover){const e=t.cover,i=bt(t.cover)?{color:t.cover}:t.cover;this.cover.load(void 0!==e.color?e:{color:i})}void 0!==t.enable&&(this.enable=t.enable)}}}class ue{constructor(){this.enable=!0,this.zIndex=0}load(t){t&&(void 0!==t.enable&&(this.enable=t.enable),void 0!==t.zIndex&&(this.zIndex=t.zIndex))}}class pe{constructor(){this.enable=!1,this.mode=[]}load(t){t&&(void 0!==t.enable&&(this.enable=t.enable),void 0!==t.mode&&(this.mode=t.mode))}}class fe{constructor(){this.selectors=[],this.enable=!1,this.mode=[],this.type="circle"}load(t){t&&(void 0!==t.selectors&&(this.selectors=t.selectors),void 0!==t.enable&&(this.enable=t.enable),void 0!==t.mode&&(this.mode=t.mode),void 0!==t.type&&(this.type=t.type))}}class ve{constructor(){this.enable=!1,this.force=2,this.smooth=10}load(t){t&&(void 0!==t.enable&&(this.enable=t.enable),void 0!==t.force&&(this.force=t.force),void 0!==t.smooth&&(this.smooth=t.smooth))}}class ye{constructor(){this.enable=!1,this.mode=[],this.parallax=new ve}load(t){t&&(void 0!==t.enable&&(this.enable=t.enable),void 0!==t.mode&&(this.mode=t.mode),this.parallax.load(t.parallax))}}class me{constructor(){this.delay=.5,this.enable=!0}load(t){void 0!==t&&(void 0!==t.delay&&(this.delay=t.delay),void 0!==t.enable&&(this.enable=t.enable))}}class ge{constructor(){this.onClick=new pe,this.onDiv=new fe,this.onHover=new ye,this.resize=new me}load(t){if(!t)return;this.onClick.load(t.onClick);const e=t.onDiv;void 0!==e&&(this.onDiv=dt(e,(t=>{const e=new fe;return e.load(t),e}))),this.onHover.load(t.onHover),this.resize.load(t.resize)}}class be{constructor(t,e){this._engine=t,this._container=e}load(t){if(!t)return;if(!this._container)return;const e=this._engine.interactors.get(this._container);if(e)for(const i of e)i.loadModeOptions&&i.loadModeOptions(this,t)}}class we{constructor(t,e){this.detectsOn="window",this.events=new ge,this.modes=new be(t,e)}load(t){if(!t)return;const e=t.detectsOn;void 0!==e&&(this.detectsOn=e),this.events.load(t.events),this.modes.load(t.modes)}}class xe{load(t){t&&(t.position&&(this.position={x:t.position.x??50,y:t.position.y??50,mode:t.position.mode??"percent"}),t.options&&(this.options=st({},t.options)))}}class _e{constructor(){this.maxWidth=1/0,this.options={},this.mode="canvas"}load(t){t&&(void 0!==t.maxWidth&&(this.maxWidth=t.maxWidth),void 0!==t.mode&&("screen"===t.mode?this.mode="screen":this.mode="canvas"),void 0!==t.options&&(this.options=st({},t.options)))}}class ke{constructor(){this.auto=!1,this.mode="any",this.value=!1}load(t){t&&(void 0!==t.auto&&(this.auto=t.auto),void 0!==t.mode&&(this.mode=t.mode),void 0!==t.value&&(this.value=t.value))}}class ze{constructor(){this.name="",this.default=new ke}load(t){t&&(void 0!==t.name&&(this.name=t.name),this.default.load(t.default),void 0!==t.options&&(this.options=st({},t.options)))}}class Me{constructor(){this.count=0,this.enable=!1,this.speed=1,this.decay=0,this.delay=0,this.sync=!1}load(t){t&&(void 0!==t.count&&(this.count=S(t.count)),void 0!==t.enable&&(this.enable=t.enable),void 0!==t.speed&&(this.speed=S(t.speed)),void 0!==t.decay&&(this.decay=S(t.decay)),void 0!==t.delay&&(this.delay=S(t.delay)),void 0!==t.sync&&(this.sync=t.sync))}}class Ce extends Me{constructor(){super(),this.mode="auto",this.startValue="random"}load(t){super.load(t),t&&(void 0!==t.mode&&(this.mode=t.mode),void 0!==t.startValue&&(this.startValue=t.startValue))}}class Pe extends Me{constructor(){super(),this.offset=0,this.sync=!0}load(t){super.load(t),t&&void 0!==t.offset&&(this.offset=S(t.offset))}}class Oe{constructor(){this.h=new Pe,this.s=new Pe,this.l=new Pe}load(t){t&&(this.h.load(t.h),this.s.load(t.s),this.l.load(t.l))}}class Se extends ce{constructor(){super(),this.animation=new Oe}static create(t,e){const i=new Se;return i.load(t),void 0!==e&&(bt(e)||kt(e)?i.load({value:e}):i.load(e)),i}load(t){if(super.load(t),!t)return;const e=t.animation;void 0!==e&&(void 0!==e.enable?this.animation.h.load(e):this.animation.load(t.animation))}}class De{constructor(){this.speed=2}load(t){t&&void 0!==t.speed&&(this.speed=t.speed)}}class Te{constructor(){this.enable=!0,this.retries=0}load(t){t&&(void 0!==t.enable&&(this.enable=t.enable),void 0!==t.retries&&(this.retries=t.retries))}}class Re{constructor(){this.value=0}load(t){t&&void 0!==t.value&&(this.value=S(t.value))}}class Ee extends Re{constructor(){super(),this.animation=new Me}load(t){if(super.load(t),!t)return;const e=t.animation;void 0!==e&&this.animation.load(e)}}class Ie extends Ee{constructor(){super(),this.animation=new Ce}load(t){super.load(t)}}class Le extends Re{constructor(){super(),this.value=1}}class Ae{constructor(){this.horizontal=new Le,this.vertical=new Le}load(t){t&&(this.horizontal.load(t.horizontal),this.vertical.load(t.vertical))}}class Fe{constructor(){this.absorb=new De,this.bounce=new Ae,this.enable=!1,this.maxSpeed=50,this.mode="bounce",this.overlap=new Te}load(t){t&&(this.absorb.load(t.absorb),this.bounce.load(t.bounce),void 0!==t.enable&&(this.enable=t.enable),void 0!==t.maxSpeed&&(this.maxSpeed=S(t.maxSpeed)),void 0!==t.mode&&(this.mode=t.mode),this.overlap.load(t.overlap))}}class Be{constructor(){this.close=!0,this.fill=!0,this.options={},this.type=[]}load(t){if(!t)return;const e=t.options;if(void 0!==e)for(const t in e){const i=e[t];i&&(this.options[t]=st(this.options[t]??{},i))}void 0!==t.close&&(this.close=t.close),void 0!==t.fill&&(this.fill=t.fill),void 0!==t.type&&(this.type=t.type)}}class qe{constructor(){this.offset=0,this.value=90}load(t){t&&(void 0!==t.offset&&(this.offset=S(t.offset)),void 0!==t.value&&(this.value=S(t.value)))}}class He{constructor(){this.distance=200,this.enable=!1,this.rotate={x:3e3,y:3e3}}load(t){if(t&&(void 0!==t.distance&&(this.distance=S(t.distance)),void 0!==t.enable&&(this.enable=t.enable),t.rotate)){const e=t.rotate.x;void 0!==e&&(this.rotate.x=e);const i=t.rotate.y;void 0!==i&&(this.rotate.y=i)}}}class Ve{constructor(){this.x=50,this.y=50,this.mode="percent",this.radius=0}load(t){t&&(void 0!==t.x&&(this.x=t.x),void 0!==t.y&&(this.y=t.y),void 0!==t.mode&&(this.mode=t.mode),void 0!==t.radius&&(this.radius=t.radius))}}class Ue{constructor(){this.acceleration=9.81,this.enable=!1,this.inverse=!1,this.maxSpeed=50}load(t){t&&(void 0!==t.acceleration&&(this.acceleration=S(t.acceleration)),void 0!==t.enable&&(this.enable=t.enable),void 0!==t.inverse&&(this.inverse=t.inverse),void 0!==t.maxSpeed&&(this.maxSpeed=S(t.maxSpeed)))}}class We{constructor(){this.clamp=!0,this.delay=new Re,this.enable=!1,this.options={}}load(t){t&&(void 0!==t.clamp&&(this.clamp=t.clamp),this.delay.load(t.delay),void 0!==t.enable&&(this.enable=t.enable),this.generator=t.generator,t.options&&(this.options=st(this.options,t.options)))}}class $e{load(t){t&&(void 0!==t.color&&(this.color=ce.create(this.color,t.color)),void 0!==t.image&&(this.image=t.image))}}class je{constructor(){this.enable=!1,this.length=10,this.fill=new $e}load(t){t&&(void 0!==t.enable&&(this.enable=t.enable),void 0!==t.fill&&this.fill.load(t.fill),void 0!==t.length&&(this.length=t.length))}}class Ge{constructor(){this.default="out"}load(t){t&&(void 0!==t.default&&(this.default=t.default),this.bottom=t.bottom??t.default,this.left=t.left??t.default,this.right=t.right??t.default,this.top=t.top??t.default)}}class Ne{constructor(){this.acceleration=0,this.enable=!1}load(t){t&&(void 0!==t.acceleration&&(this.acceleration=S(t.acceleration)),void 0!==t.enable&&(this.enable=t.enable),t.position&&(this.position=st({},t.position)))}}class Xe{constructor(){this.angle=new qe,this.attract=new He,this.center=new Ve,this.decay=0,this.distance={},this.direction="none",this.drift=0,this.enable=!1,this.gravity=new Ue,this.path=new We,this.outModes=new Ge,this.random=!1,this.size=!1,this.speed=2,this.spin=new Ne,this.straight=!1,this.trail=new je,this.vibrate=!1,this.warp=!1}load(t){if(!t)return;this.angle.load(wt(t.angle)?{value:t.angle}:t.angle),this.attract.load(t.attract),this.center.load(t.center),void 0!==t.decay&&(this.decay=S(t.decay)),void 0!==t.direction&&(this.direction=t.direction),void 0!==t.distance&&(this.distance=wt(t.distance)?{horizontal:t.distance,vertical:t.distance}:{...t.distance}),void 0!==t.drift&&(this.drift=S(t.drift)),void 0!==t.enable&&(this.enable=t.enable),this.gravity.load(t.gravity);const e=t.outModes;void 0!==e&&(_t(e)?this.outModes.load(e):this.outModes.load({default:e})),this.path.load(t.path),void 0!==t.random&&(this.random=t.random),void 0!==t.size&&(this.size=t.size),void 0!==t.speed&&(this.speed=S(t.speed)),this.spin.load(t.spin),void 0!==t.straight&&(this.straight=t.straight),this.trail.load(t.trail),void 0!==t.vibrate&&(this.vibrate=t.vibrate),void 0!==t.warp&&(this.warp=t.warp)}}class Ye extends Ce{constructor(){super(),this.destroy="none",this.speed=2}load(t){super.load(t),t&&void 0!==t.destroy&&(this.destroy=t.destroy)}}class Ze extends Ie{constructor(){super(),this.animation=new Ye,this.value=1}load(t){if(!t)return;super.load(t);const e=t.animation;void 0!==e&&this.animation.load(e)}}class Qe{constructor(){this.enable=!1,this.width=1920,this.height=1080}load(t){if(!t)return;void 0!==t.enable&&(this.enable=t.enable);const e=t.width;void 0!==e&&(this.width=e);const i=t.height;void 0!==i&&(this.height=i)}}class Je{constructor(){this.mode="delete",this.value=0}load(t){t&&(void 0!==t.mode&&(this.mode=t.mode),void 0!==t.value&&(this.value=t.value))}}class Ke{constructor(){this.density=new Qe,this.limit=new Je,this.value=0}load(t){t&&(this.density.load(t.density),this.limit.load(t.limit),void 0!==t.value&&(this.value=t.value))}}class ti{constructor(){this.blur=0,this.color=new ce,this.enable=!1,this.offset={x:0,y:0},this.color.value="#000"}load(t){t&&(void 0!==t.blur&&(this.blur=t.blur),this.color=ce.create(this.color,t.color),void 0!==t.enable&&(this.enable=t.enable),void 0!==t.offset&&(void 0!==t.offset.x&&(this.offset.x=t.offset.x),void 0!==t.offset.y&&(this.offset.y=t.offset.y)))}}class ei{constructor(){this.close=!0,this.fill=!0,this.options={},this.type="circle"}load(t){if(!t)return;const e=t.options;if(void 0!==e)for(const t in e){const i=e[t];i&&(this.options[t]=st(this.options[t]??{},i))}void 0!==t.close&&(this.close=t.close),void 0!==t.fill&&(this.fill=t.fill),void 0!==t.type&&(this.type=t.type)}}class ii extends Ce{constructor(){super(),this.destroy="none",this.speed=5}load(t){super.load(t),t&&void 0!==t.destroy&&(this.destroy=t.destroy)}}class si extends Ie{constructor(){super(),this.animation=new ii,this.value=3}load(t){if(super.load(t),!t)return;const e=t.animation;void 0!==e&&this.animation.load(e)}}class oi{constructor(){this.width=0}load(t){t&&(void 0!==t.color&&(this.color=Se.create(this.color,t.color)),void 0!==t.width&&(this.width=S(t.width)),void 0!==t.opacity&&(this.opacity=S(t.opacity)))}}class ni extends Re{constructor(){super(),this.opacityRate=1,this.sizeRate=1,this.velocityRate=1}load(t){super.load(t),t&&(void 0!==t.opacityRate&&(this.opacityRate=t.opacityRate),void 0!==t.sizeRate&&(this.sizeRate=t.sizeRate),void 0!==t.velocityRate&&(this.velocityRate=t.velocityRate))}}class ai{constructor(t,e){this._engine=t,this._container=e,this.bounce=new Ae,this.collisions=new Fe,this.color=new Se,this.color.value="#fff",this.effect=new Be,this.groups={},this.move=new Xe,this.number=new Ke,this.opacity=new Ze,this.reduceDuplicates=!1,this.shadow=new ti,this.shape=new ei,this.size=new si,this.stroke=new oi,this.zIndex=new ni}load(t){if(!t)return;if(void 0!==t.groups)for(const e of Object.keys(t.groups)){if(!Object.hasOwn(t.groups,e))continue;const i=t.groups[e];void 0!==i&&(this.groups[e]=st(this.groups[e]??{},i))}void 0!==t.reduceDuplicates&&(this.reduceDuplicates=t.reduceDuplicates),this.bounce.load(t.bounce),this.color.load(Se.create(this.color,t.color)),this.effect.load(t.effect),this.move.load(t.move),this.number.load(t.number),this.opacity.load(t.opacity),this.shape.load(t.shape),this.size.load(t.size),this.shadow.load(t.shadow),this.zIndex.load(t.zIndex),this.collisions.load(t.collisions),void 0!==t.interactivity&&(this.interactivity=st({},t.interactivity));const e=t.stroke;if(e&&(this.stroke=dt(e,(t=>{const e=new oi;return e.load(t),e}))),this._container){const e=this._engine.updaters.get(this._container);if(e)for(const i of e)i.loadOptions&&i.loadOptions(this,t);const i=this._engine.interactors.get(this._container);if(i)for(const e of i)e.loadParticlesOptions&&e.loadParticlesOptions(this,t)}}}function ri(t,...e){for(const i of e)t.load(i)}function ci(t,e,...i){const s=new ai(t,e);return ri(s,...i),s}class li{constructor(t,e){this._findDefaultTheme=t=>this.themes.find((e=>e.default.value&&e.default.mode===t))??this.themes.find((t=>t.default.value&&"any"===t.default.mode)),this._importPreset=t=>{this.load(this._engine.getPreset(t))},this._engine=t,this._container=e,this.autoPlay=!0,this.background=new le,this.backgroundMask=new de,this.clear=!0,this.defaultThemes={},this.delay=0,this.fullScreen=new ue,this.detectRetina=!0,this.duration=0,this.fpsLimit=120,this.interactivity=new we(t,e),this.manualParticles=[],this.particles=ci(this._engine,this._container),this.pauseOnBlur=!0,this.pauseOnOutsideViewport=!0,this.responsive=[],this.smooth=!1,this.style={},this.themes=[],this.zLayers=100}load(t){if(!t)return;void 0!==t.preset&&dt(t.preset,(t=>this._importPreset(t))),void 0!==t.autoPlay&&(this.autoPlay=t.autoPlay),void 0!==t.clear&&(this.clear=t.clear),void 0!==t.name&&(this.name=t.name),void 0!==t.delay&&(this.delay=S(t.delay));const e=t.detectRetina;void 0!==e&&(this.detectRetina=e),void 0!==t.duration&&(this.duration=S(t.duration));const i=t.fpsLimit;void 0!==i&&(this.fpsLimit=i),void 0!==t.pauseOnBlur&&(this.pauseOnBlur=t.pauseOnBlur),void 0!==t.pauseOnOutsideViewport&&(this.pauseOnOutsideViewport=t.pauseOnOutsideViewport),void 0!==t.zLayers&&(this.zLayers=t.zLayers),this.background.load(t.background);const s=t.fullScreen;gt(s)?this.fullScreen.enable=s:this.fullScreen.load(s),this.backgroundMask.load(t.backgroundMask),this.interactivity.load(t.interactivity),t.manualParticles&&(this.manualParticles=t.manualParticles.map((t=>{const e=new xe;return e.load(t),e}))),this.particles.load(t.particles),this.style=st(this.style,t.style),this._engine.loadOptions(this,t),void 0!==t.smooth&&(this.smooth=t.smooth);const o=this._engine.interactors.get(this._container);if(o)for(const e of o)e.loadOptions&&e.loadOptions(this,t);if(void 0!==t.responsive)for(const e of t.responsive){const t=new _e;t.load(e),this.responsive.push(t)}if(this.responsive.sort(((t,e)=>t.maxWidth-e.maxWidth)),void 0!==t.themes)for(const e of t.themes){const t=this.themes.find((t=>t.name===e.name));if(t)t.load(e);else{const t=new ze;t.load(e),this.themes.push(t)}}this.defaultThemes.dark=this._findDefaultTheme("dark")?.name,this.defaultThemes.light=this._findDefaultTheme("light")?.name}setResponsive(t,e,i){this.load(i);const s=this.responsive.find((i=>"screen"===i.mode&&screen?i.maxWidth>screen.availWidth:i.maxWidth*e>t));return this.load(s?.options),s?.maxWidth}setTheme(t){if(t){const e=this.themes.find((e=>e.name===t));e&&this.load(e.options)}else{const t=N("(prefers-color-scheme: dark)"),e=t&&t.matches,i=this._findDefaultTheme(e?"dark":"light");i&&this.load(i.options)}}}class hi{constructor(t,e){this.container=e,this._engine=t,this._interactors=t.getInteractors(this.container,!0),this._externalInteractors=[],this._particleInteractors=[]}async externalInteract(t){for(const e of this._externalInteractors)e.isEnabled()&&await e.interact(t)}handleClickMode(t){for(const e of this._externalInteractors)e.handleClickMode&&e.handleClickMode(t)}init(){this._externalInteractors=[],this._particleInteractors=[];for(const t of this._interactors){switch(t.type){case"external":this._externalInteractors.push(t);break;case"particles":this._particleInteractors.push(t)}t.init()}}async particlesInteract(t,e){for(const i of this._externalInteractors)i.clear(t,e);for(const i of this._particleInteractors)i.isEnabled(t)&&await i.interact(t,e)}async reset(t){for(const e of this._externalInteractors)e.isEnabled()&&e.reset(t);for(const e of this._particleInteractors)e.isEnabled(t)&&e.reset(t)}}function di(t){if(!Z(t.outMode,t.checkModes))return;const e=2*t.radius;t.coord>t.maxCoord-e?t.setCb(-t.radius):t.coord{for(const[,s]of t.plugins){const t=void 0!==s.particlePosition?s.particlePosition(e,this):void 0;if(t)return v.create(t.x,t.y,i)}const o=B({size:t.canvas.size,position:e}),n=v.create(o.x,o.y,i),a=this.getRadius(),r=this.options.move.outModes,c=e=>{di({outMode:e,checkModes:["bounce","bounce-horizontal"],coord:n.x,maxCoord:t.canvas.size.width,setCb:t=>n.x+=t,radius:a})},l=e=>{di({outMode:e,checkModes:["bounce","bounce-vertical"],coord:n.y,maxCoord:t.canvas.size.height,setCb:t=>n.y+=t,radius:a})};return c(r.left??r.default),c(r.right??r.default),l(r.top??r.default),l(r.bottom??r.default),this._checkOverlap(n,s)?this._calcPosition(t,void 0,i,s+1):n},this._calculateVelocity=()=>{const t=E(this.direction).copy(),e=this.options.move;if("inside"===e.direction||"outside"===e.direction)return t;const i=Math.PI/180*C(e.angle.value),s=Math.PI/180*C(e.angle.offset),o={left:s-.5*i,right:s+.5*i};return e.straight||(t.angle+=M(S(o.left,o.right))),e.random&&"number"==typeof e.speed&&(t.length*=_()),t},this._checkOverlap=(t,e=0)=>{const i=this.options.collisions,s=this.getRadius();if(!i.enable)return!1;const o=i.overlap;if(o.enable)return!1;const n=o.retries;if(n>=0&&e>n)throw new Error(`${f} particle is overlapping and can't be placed`);return!!this.container.particles.find((e=>T(t,e.position){if(!t||!this.roll||!this.backColor&&!this.roll.alter)return t;const e=this.roll.horizontal&&this.roll.vertical?2:1,i=this.roll.horizontal?.5*Math.PI:0;return Math.floor(((this.roll.angle??0)+i)/(Math.PI/e))%2?this.backColor?this.backColor:this.roll.alter?se(t,this.roll.alter.type,this.roll.alter.value):t:t},this._initPosition=t=>{const e=this.container,i=C(this.options.zIndex.value);this.position=this._calcPosition(e,t,k(i,0,e.zLayers)),this.initialPosition=this.position.copy();const s=e.canvas.size;switch(this.moveCenter={...yt(this.options.move.center,s),radius:this.options.move.center.radius??0,mode:this.options.move.center.mode??"percent"},this.direction=R(this.options.move.direction,this.position,this.moveCenter),this.options.move.direction){case"inside":this.outType="inside";break;case"outside":this.outType="outside"}this.offset=y.origin},this._engine=t,this.init(e,s,o,n)}destroy(t){if(this.unbreakable||this.destroyed)return;this.destroyed=!0,this.bubble.inRange=!1,this.slow.inRange=!1;const e=this.container,i=this.pathGenerator,s=e.shapeDrawers.get(this.shape);s&&s.particleDestroy&&s.particleDestroy(this);for(const[,i]of e.plugins)i.particleDestroyed&&i.particleDestroyed(this,t);for(const i of e.particles.updaters)i.particleDestroyed&&i.particleDestroyed(this,t);i&&i.reset(this),this._engine.dispatchEvent("particleDestroyed",{container:this.container,data:{particle:this}})}draw(t){const e=this.container,i=e.canvas;for(const[,s]of e.plugins)i.drawParticlePlugin(s,this,t);i.drawParticle(this,t)}getFillColor(){return this._getRollColor(this.bubble.color??$t(this.color))}getMass(){return this.getRadius()**2*Math.PI*.5}getPosition(){return{x:this.position.x+this.offset.x,y:this.position.y+this.offset.y,z:this.position.z}}getRadius(){return this.bubble.radius??this.size.value}getStrokeColor(){return this._getRollColor(this.bubble.color??$t(this.strokeColor))}init(t,e,i,s){const o=this.container,n=this._engine;this.id=t,this.group=s,this.effectClose=!0,this.effectFill=!0,this.shapeClose=!0,this.shapeFill=!0,this.pathRotation=!1,this.lastPathTime=0,this.destroyed=!1,this.unbreakable=!1,this.rotation=0,this.misplaced=!1,this.retina={maxDistance:{}},this.outType="normal",this.ignoresResizeRatio=!0;const a=o.retina.pixelRatio,r=o.actualOptions,c=ci(this._engine,o,r.particles),l=c.effect.type,h=c.shape.type,{reduceDuplicates:d}=c;this.effect=ut(l,this.id,d),this.shape=ut(h,this.id,d);const u=c.effect,p=c.shape;if(i){if(i.effect&&i.effect.type){const t=ut(i.effect.type,this.id,d);t&&(this.effect=t,u.load(i.effect))}if(i.shape&&i.shape.type){const t=ut(i.shape.type,this.id,d);t&&(this.shape=t,p.load(i.shape))}}this.effectData=function(t,e,i,s){const o=e.options[t];if(o)return st({close:e.close,fill:e.fill},ut(o,i,s))}(this.effect,u,this.id,d),this.shapeData=function(t,e,i,s){const o=e.options[t];if(o)return st({close:e.close,fill:e.fill},ut(o,i,s))}(this.shape,p,this.id,d),c.load(i);const f=this.effectData;f&&c.load(f.particles);const v=this.shapeData;v&&c.load(v.particles);const y=new we(n,o);y.load(o.actualOptions.interactivity),y.load(c.interactivity),this.interactivity=y,this.effectFill=f?.fill??c.effect.fill,this.effectClose=f?.close??c.effect.close,this.shapeFill=v?.fill??c.shape.fill,this.shapeClose=v?.close??c.shape.close,this.options=c;const m=this.options.move.path;this.pathDelay=1e3*C(m.delay.value),m.generator&&(this.pathGenerator=this._engine.getPathGenerator(m.generator),this.pathGenerator&&o.addPath(m.generator,this.pathGenerator)&&this.pathGenerator.init(o)),o.retina.initParticle(this),this.size=ft(this.options.size,a),this.bubble={inRange:!1},this.slow={inRange:!1,factor:1},this._initPosition(e),this.initialVelocity=this._calculateVelocity(),this.velocity=this.initialVelocity.copy(),this.moveDecay=1-C(this.options.move.decay);const g=o.particles;g.setLastZIndex(this.position.z),this.zIndexFactor=this.position.z/o.zLayers,this.sides=24;let b=o.effectDrawers.get(this.effect);b||(b=this._engine.getEffectDrawer(this.effect),b&&o.effectDrawers.set(this.effect,b)),b&&b.loadEffect&&b.loadEffect(this);let w=o.shapeDrawers.get(this.shape);w||(w=this._engine.getShapeDrawer(this.shape),w&&o.shapeDrawers.set(this.shape,w)),w&&w.loadShape&&w.loadShape(this);const x=w?.getSidesCount;x&&(this.sides=x(this)),this.spawning=!1,this.shadowColor=St(this.options.shadow.color);for(const t of g.updaters)t.init(this);for(const t of g.movers)t.init&&t.init(this);b&&b.particleInit&&b.particleInit(o,this),w&&w.particleInit&&w.particleInit(o,this);for(const[,t]of o.plugins)t.particleCreated&&t.particleCreated(this)}isInsideCanvas(){const t=this.getRadius(),e=this.container.canvas.size,i=this.position;return i.x>=-t&&i.y>=-t&&i.y<=e.height+t&&i.x<=e.width+t}isVisible(){return!this.destroyed&&!this.spawning&&this.isInsideCanvas()}reset(){for(const t of this.container.particles.updaters)t.reset&&t.reset(this)}}class pi{constructor(t,e){this.position=t,this.particle=e}}class fi{constructor(t,e){this.position={x:t,y:e}}}class vi extends fi{constructor(t,e,i,s){super(t,e),this.size={height:s,width:i}}contains(t){const e=this.size.width,i=this.size.height,s=this.position;return t.x>=s.x&&t.x<=s.x+e&&t.y>=s.y&&t.y<=s.y+i}intersects(t){t instanceof yi&&t.intersects(this);const e=this.size.width,i=this.size.height,s=this.position,o=t.position,n=t instanceof vi?t.size:{width:0,height:0},a=n.width,r=n.height;return o.xs.x&&o.ys.y}}class yi extends fi{constructor(t,e,i){super(t,e),this.radius=i}contains(t){return T(t,this.position)<=this.radius}intersects(t){const e=this.position,i=t.position,s=Math.abs(i.x-e.x),o=Math.abs(i.y-e.y),n=this.radius;if(t instanceof yi){return n+t.radius>Math.sqrt(s**2+o**2)}if(t instanceof vi){const{width:e,height:i}=t.size;return Math.pow(s-e,2)+Math.pow(o-i,2)<=n**2||s<=n+e&&o<=n+i||s<=e||o<=i}return!1}}class mi{constructor(t,e){this.rectangle=t,this.capacity=e,this._subdivide=()=>{const{x:t,y:e}=this.rectangle.position,{width:i,height:s}=this.rectangle.size,{capacity:o}=this;for(let n=0;n<4;n++)this._subs.push(new mi(new vi(t+.5*i*(n%2),e+.5*s*(Math.round(.5*n)-n%2),.5*i,.5*s),o));this._divided=!0},this._points=[],this._divided=!1,this._subs=[]}insert(t){return!!this.rectangle.contains(t.position)&&(this._points.lengthe.insert(t)))))}query(t,e,i){const s=i||[];if(!t.intersects(this.rectangle))return[];for(const i of this._points)!t.contains(i.position)&&T(t.position,i.position)>i.particle.getRadius()&&(!e||e(i.particle))||s.push(i.particle);if(this._divided)for(const i of this._subs)i.query(t,e,s);return s}queryCircle(t,e,i){return this.query(new yi(t.x,t.y,e),i)}queryRectangle(t,e,i){return this.query(new vi(t.x,t.y,e.width,e.height),i)}}const gi=t=>{const{height:e,width:i}=t;return new vi(-.25*i,-.25*e,1.5*i,1.5*e)};class bi{constructor(t,e){this._addToPool=(...t)=>{for(const e of t)this._pool.push(e)},this._applyDensity=(t,e,i)=>{const s=t.number;if(!t.number.density?.enable)return void(void 0===i?this._limit=s.limit.value:s.limit&&this._groupLimits.set(i,s.limit.value));const o=this._initDensityFactor(s.density),n=s.value,a=s.limit.value>0?s.limit.value:n,r=Math.min(n,a)*o+e,c=Math.min(this.count,this.filter((t=>t.group===i)).length);void 0===i?this._limit=s.limit.value*o:this._groupLimits.set(i,s.limit.value*o),cr&&this.removeQuantity(c-r,i)},this._initDensityFactor=t=>{const e=this._container;if(!e.canvas.element||!t.enable)return 1;const i=e.canvas.element,s=e.retina.pixelRatio;return i.width*i.height/(t.height*t.width*s**2)},this._pushParticle=(t,e,i,s)=>{try{let o=this._pool.pop();o?o.init(this._nextId,t,e,i):o=new ui(this._engine,this._nextId,this._container,t,e,i);let n=!0;if(s&&(n=s(o)),!n)return;return this._array.push(o),this._zArray.push(o),this._nextId++,this._engine.dispatchEvent("particleAdded",{container:this._container,data:{particle:o}}),o}catch(t){return void W().warning(`${f} adding particle: ${t}`)}},this._removeParticle=(t,e,i)=>{const s=this._array[t];if(!s||s.group!==e)return!1;const o=this._zArray.indexOf(s);return this._array.splice(t,1),this._zArray.splice(o,1),s.destroy(i),this._engine.dispatchEvent("particleRemoved",{container:this._container,data:{particle:s}}),this._addToPool(s),!0},this._engine=t,this._container=e,this._nextId=0,this._array=[],this._zArray=[],this._pool=[],this._limit=0,this._groupLimits=new Map,this._needsSort=!1,this._lastZIndex=0,this._interactionManager=new hi(t,e);const i=e.canvas.size;this.quadTree=new mi(gi(i),4),this.movers=this._engine.getMovers(e,!0),this.updaters=this._engine.getUpdaters(e,!0)}get count(){return this._array.length}addManualParticles(){const t=this._container,e=t.actualOptions;for(const i of e.manualParticles)this.addParticle(i.position?yt(i.position,t.canvas.size):void 0,i.options)}addParticle(t,e,i,s){const o=this._container.actualOptions.particles.number.limit,n=void 0===i?this._limit:this._groupLimits.get(i)??this._limit,a=this.count;if(n>0)if("delete"===o.mode){const t=a+1-n;t>0&&this.removeQuantity(t)}else if("wait"===o.mode&&a>=n)return;return this._pushParticle(t,e,i,s)}clear(){this._array=[],this._zArray=[]}destroy(){this._array=[],this._zArray=[],this.movers=[],this.updaters=[]}async draw(t){const e=this._container,i=e.canvas;i.clear(),await this.update(t);for(const[,s]of e.plugins)i.drawPlugin(s,t);for(const e of this._zArray)e.draw(t)}filter(t){return this._array.filter(t)}find(t){return this._array.find(t)}get(t){return this._array[t]}handleClickMode(t){this._interactionManager.handleClickMode(t)}init(){const t=this._container,e=t.actualOptions;this._lastZIndex=0,this._needsSort=!1;let i=!1;this.updaters=this._engine.getUpdaters(t,!0),this._interactionManager.init();for(const[,e]of t.plugins)if(void 0!==e.particlesInitialization&&(i=e.particlesInitialization()),i)break;this._interactionManager.init();for(const[,e]of t.pathGenerators)e.init(t);if(this.addManualParticles(),!i){const t=e.particles,i=t.groups;for(const e in i){const s=i[e];for(let i=this.count,o=0;othis.count)return;let o=0;for(let n=t;o!i.has(t);this._array=this.filter(t),this._zArray=this._zArray.filter(t);for(const t of i)this._engine.dispatchEvent("particleRemoved",{container:this._container,data:{particle:t}});this._addToPool(...i)}await this._interactionManager.externalInteract(t);for(const e of this._array){for(const i of this.updaters)i.update(e,t);e.destroyed||e.spawning||await this._interactionManager.particlesInteract(e,t)}if(delete this._resizeFactor,this._needsSort){const t=this._zArray;t.sort(((t,e)=>e.position.z-t.position.z||t.id-e.id)),this._lastZIndex=t[t.length-1].position.z,this._needsSort=!1}}}class wi{constructor(t){this.container=t,this.pixelRatio=1,this.reduceFactor=1}init(){const t=this.container,e=t.actualOptions;this.pixelRatio=!e.detectRetina||j()?1:window.devicePixelRatio,this.reduceFactor=1;const i=this.pixelRatio,s=t.canvas;if(s.element){const t=s.element;s.size.width=t.offsetWidth*i,s.size.height=t.offsetHeight*i}const o=e.particles,n=o.move;this.maxSpeed=C(n.gravity.maxSpeed)*i,this.sizeAnimationSpeed=C(o.size.animation.speed)*i}initParticle(t){const e=t.options,i=this.pixelRatio,s=e.move,o=s.distance,n=t.retina;n.moveDrift=C(s.drift)*i,n.moveSpeed=C(s.speed)*i,n.sizeAnimationSpeed=C(e.size.animation.speed)*i;const a=n.maxDistance;a.horizontal=void 0!==o.horizontal?o.horizontal*i:void 0,a.vertical=void 0!==o.vertical?o.vertical*i:void 0,n.maxSpeed=C(s.gravity.maxSpeed)*i}}function xi(t){return t&&!t.destroyed}function _i(t,e,...i){const s=new li(t,e);return ri(s,...i),s}class ki{constructor(t,e,i){this._intersectionManager=t=>{if(xi(this)&&this.actualOptions.pauseOnOutsideViewport)for(const e of t)e.target===this.interactivity.element&&(e.isIntersecting?this.play:this.pause)()},this._nextFrame=async t=>{try{if(!this._smooth&&void 0!==this._lastFrameTime&&t1e3)return void this.draw(!1);if(await this.particles.draw(e),!this.alive())return void this.destroy();this.getAnimationStatus()&&this.draw(!1)}catch(t){W().error(`${f} in animation loop`,t)}},this._engine=t,this.id=Symbol(e),this.fpsLimit=120,this._smooth=!1,this._delay=0,this._duration=0,this._lifeTime=0,this._firstStart=!0,this.started=!1,this.destroyed=!1,this._paused=!0,this._lastFrameTime=0,this.zLayers=100,this.pageHidden=!1,this._sourceOptions=i,this._initialSourceOptions=i,this.retina=new wi(this),this.canvas=new ne(this),this.particles=new bi(this._engine,this),this.pathGenerators=new Map,this.interactivity={mouse:{clicking:!1,inside:!1}},this.plugins=new Map,this.effectDrawers=new Map,this.shapeDrawers=new Map,this._options=_i(this._engine,this),this.actualOptions=_i(this._engine,this),this._eventListeners=new re(this),this._intersectionObserver=X((t=>this._intersectionManager(t))),this._engine.dispatchEvent("containerBuilt",{container:this})}get options(){return this._options}get sourceOptions(){return this._sourceOptions}addClickHandler(t){if(!xi(this))return;const e=this.interactivity.element;if(!e)return;const i=(e,i,s)=>{if(!xi(this))return;const o=this.retina.pixelRatio,n={x:i.x*o,y:i.y*o},a=this.particles.quadTree.queryCircle(n,s*o);t(e,a)};let s=!1,o=!1;e.addEventListener("click",(t=>{if(!xi(this))return;const e=t,s={x:e.offsetX||e.clientX,y:e.offsetY||e.clientY};i(t,s,1)})),e.addEventListener("touchstart",(()=>{xi(this)&&(s=!0,o=!1)})),e.addEventListener("touchmove",(()=>{xi(this)&&(o=!0)})),e.addEventListener("touchend",(t=>{if(xi(this)){if(s&&!o){const e=t;let s=e.touches[e.touches.length-1];if(!s&&(s=e.changedTouches[e.changedTouches.length-1],!s))return;const o=this.canvas.element,n=o?o.getBoundingClientRect():void 0,a={x:s.clientX-(n?n.left:0),y:s.clientY-(n?n.top:0)};i(t,a,Math.max(s.radiusX,s.radiusY))}s=!1,o=!1}})),e.addEventListener("touchcancel",(()=>{xi(this)&&(s=!1,o=!1)}))}addLifeTime(t){this._lifeTime+=t}addPath(t,e,i=!1){return!(!xi(this)||!i&&this.pathGenerators.has(t))&&(this.pathGenerators.set(t,e),!0)}alive(){return!this._duration||this._lifeTime<=this._duration}destroy(){if(!xi(this))return;this.stop(),this.particles.destroy(),this.canvas.destroy();for(const[,t]of this.effectDrawers)t.destroy&&t.destroy(this);for(const[,t]of this.shapeDrawers)t.destroy&&t.destroy(this);for(const t of this.effectDrawers.keys())this.effectDrawers.delete(t);for(const t of this.shapeDrawers.keys())this.shapeDrawers.delete(t);this._engine.clearPlugins(this),this.destroyed=!0;const t=this._engine.dom(),e=t.findIndex((t=>t===this));e>=0&&t.splice(e,1),this._engine.dispatchEvent("containerDestroyed",{container:this})}draw(t){if(!xi(this))return;let e=t;this._drawAnimationFrame=requestAnimationFrame((async t=>{e&&(this._lastFrameTime=void 0,e=!1),await this._nextFrame(t)}))}async export(t,e={}){for(const[,i]of this.plugins){if(!i.export)continue;const s=await i.export(t,e);if(s.supported)return s.blob}W().error(`${f} - Export plugin with type ${t} not found`)}getAnimationStatus(){return!this._paused&&!this.pageHidden&&xi(this)}handleClickMode(t){if(xi(this)){this.particles.handleClickMode(t);for(const[,e]of this.plugins)e.handleClickMode&&e.handleClickMode(t)}}async init(){if(!xi(this))return;const t=this._engine.getSupportedEffects();for(const e of t){const t=this._engine.getEffectDrawer(e);t&&this.effectDrawers.set(e,t)}const e=this._engine.getSupportedShapes();for(const t of e){const e=this._engine.getShapeDrawer(t);e&&this.shapeDrawers.set(t,e)}this._options=_i(this._engine,this,this._initialSourceOptions,this.sourceOptions),this.actualOptions=_i(this._engine,this,this._options);const i=this._engine.getAvailablePlugins(this);for(const[t,e]of i)this.plugins.set(t,e);this.retina.init(),await this.canvas.init(),this.updateActualOptions(),this.canvas.initBackground(),this.canvas.resize(),this.zLayers=this.actualOptions.zLayers,this._duration=1e3*C(this.actualOptions.duration),this._delay=1e3*C(this.actualOptions.delay),this._lifeTime=0,this.fpsLimit=this.actualOptions.fpsLimit>0?this.actualOptions.fpsLimit:120,this._smooth=this.actualOptions.smooth;for(const[,t]of this.effectDrawers)t.init&&await t.init(this);for(const[,t]of this.shapeDrawers)t.init&&await t.init(this);for(const[,t]of this.plugins)t.init&&await t.init();this._engine.dispatchEvent("containerInit",{container:this}),this.particles.init(),this.particles.setDensity();for(const[,t]of this.plugins)t.particlesSetup&&t.particlesSetup();this._engine.dispatchEvent("particlesSetup",{container:this})}async loadTheme(t){xi(this)&&(this._currentTheme=t,await this.refresh())}pause(){if(xi(this)&&(void 0!==this._drawAnimationFrame&&(cancelAnimationFrame(this._drawAnimationFrame),delete this._drawAnimationFrame),!this._paused)){for(const[,t]of this.plugins)t.pause&&t.pause();this.pageHidden||(this._paused=!0),this._engine.dispatchEvent("containerPaused",{container:this})}}play(t){if(!xi(this))return;const e=this._paused||t;if(!this._firstStart||this.actualOptions.autoPlay){if(this._paused&&(this._paused=!1),e)for(const[,t]of this.plugins)t.play&&t.play();this._engine.dispatchEvent("containerPlay",{container:this}),this.draw(e||!1)}else this._firstStart=!1}async refresh(){if(xi(this))return this.stop(),this.start()}async reset(){if(xi(this))return this._initialSourceOptions=void 0,this._options=_i(this._engine,this),this.actualOptions=_i(this._engine,this,this._options),this.refresh()}async start(){xi(this)&&!this.started&&(await this.init(),this.started=!0,await new Promise((t=>{this._delayTimeout=setTimeout((async()=>{this._eventListeners.addListeners(),this.interactivity.element instanceof HTMLElement&&this._intersectionObserver&&this._intersectionObserver.observe(this.interactivity.element);for(const[,t]of this.plugins)t.start&&await t.start();this._engine.dispatchEvent("containerStarted",{container:this}),this.play(),t()}),this._delay)})))}stop(){if(xi(this)&&this.started){this._delayTimeout&&(clearTimeout(this._delayTimeout),delete this._delayTimeout),this._firstStart=!0,this.started=!1,this._eventListeners.removeListeners(),this.pause(),this.particles.clear(),this.canvas.stop(),this.interactivity.element instanceof HTMLElement&&this._intersectionObserver&&this._intersectionObserver.unobserve(this.interactivity.element);for(const[,t]of this.plugins)t.stop&&t.stop();for(const t of this.plugins.keys())this.plugins.delete(t);this._sourceOptions=this._options,this._engine.dispatchEvent("containerStopped",{container:this})}}updateActualOptions(){this.actualOptions.responsive=[];const t=this.actualOptions.setResponsive(this.canvas.size.width,this.retina.pixelRatio,this._options);return this.actualOptions.setTheme(this._currentTheme),this._responsiveMaxWidth!==t&&(this._responsiveMaxWidth=t,!0)}}class zi{constructor(){this._listeners=new Map}addEventListener(t,e){this.removeEventListener(t,e);let i=this._listeners.get(t);i||(i=[],this._listeners.set(t,i)),i.push(e)}dispatchEvent(t,e){const i=this._listeners.get(t);i&&i.forEach((t=>t(e)))}hasEventListener(t){return!!this._listeners.get(t)}removeAllEventListeners(t){t?this._listeners.delete(t):this._listeners=new Map}removeEventListener(t,e){const i=this._listeners.get(t);if(!i)return;const s=i.length,o=i.indexOf(e);o<0||(1===s?this._listeners.delete(t):i.splice(o,1))}}function Mi(t,e,i,s=!1){let o=e.get(t);return o&&!s||(o=[...i.values()].map((e=>e(t))),e.set(t,o)),o}class Ci{constructor(){this._configs=new Map,this._domArray=[],this._eventDispatcher=new zi,this._initialized=!1,this.plugins=[],this._initializers={interactors:new Map,movers:new Map,updaters:new Map},this.interactors=new Map,this.movers=new Map,this.updaters=new Map,this.presets=new Map,this.effectDrawers=new Map,this.shapeDrawers=new Map,this.pathGenerators=new Map}get configs(){const t={};for(const[e,i]of this._configs)t[e]=i;return t}get version(){return"3.0.2"}addConfig(t){const e=t.name??"default";this._configs.set(e,t),this._eventDispatcher.dispatchEvent("configAdded",{data:{name:e,config:t}})}async addEffect(t,e,i=!0){dt(t,(t=>{!this.getEffectDrawer(t)&&this.effectDrawers.set(t,e)})),await this.refresh(i)}addEventListener(t,e){this._eventDispatcher.addEventListener(t,e)}async addInteractor(t,e,i=!0){this._initializers.interactors.set(t,e),await this.refresh(i)}async addMover(t,e,i=!0){this._initializers.movers.set(t,e),await this.refresh(i)}async addParticleUpdater(t,e,i=!0){this._initializers.updaters.set(t,e),await this.refresh(i)}async addPathGenerator(t,e,i=!0){!this.getPathGenerator(t)&&this.pathGenerators.set(t,e),await this.refresh(i)}async addPlugin(t,e=!0){!this.getPlugin(t.id)&&this.plugins.push(t),await this.refresh(e)}async addPreset(t,e,i=!1,s=!0){(i||!this.getPreset(t))&&this.presets.set(t,e),await this.refresh(s)}async addShape(t,e,i=!0){dt(t,(t=>{!this.getShapeDrawer(t)&&this.shapeDrawers.set(t,e)})),await this.refresh(i)}clearPlugins(t){this.updaters.delete(t),this.movers.delete(t),this.interactors.delete(t)}dispatchEvent(t,e){this._eventDispatcher.dispatchEvent(t,e)}dom(){return this._domArray}domItem(t){const e=this.dom(),i=e[t];if(i&&!i.destroyed)return i;e.splice(t,1)}getAvailablePlugins(t){const e=new Map;for(const i of this.plugins)i.needsPlugin(t.actualOptions)&&e.set(i.id,i.getPlugin(t));return e}getEffectDrawer(t){return this.effectDrawers.get(t)}getInteractors(t,e=!1){return Mi(t,this.interactors,this._initializers.interactors,e)}getMovers(t,e=!1){return Mi(t,this.movers,this._initializers.movers,e)}getPathGenerator(t){return this.pathGenerators.get(t)}getPlugin(t){return this.plugins.find((e=>e.id===t))}getPreset(t){return this.presets.get(t)}getShapeDrawer(t){return this.shapeDrawers.get(t)}getSupportedEffects(){return this.effectDrawers.keys()}getSupportedShapes(){return this.shapeDrawers.keys()}getUpdaters(t,e=!1){return Mi(t,this.updaters,this._initializers.updaters,e)}init(){this._initialized||(this._initialized=!0)}async load(t){const e=t.id??`tsparticles${Math.floor(1e4*_())}`,{index:s,url:o}=t,n=o?await async function(t){const e=ut(t.url,t.index);if(!e)return t.fallback;const i=await fetch(e);return i.ok?i.json():(W().error(`${f} ${i.status} while retrieving config file`),t.fallback)}({fallback:t.options,url:o,index:s}):t.options;let a=t.element??document.getElementById(e);a||(a=document.createElement("div"),a.id=e,document.body.append(a));const r=ut(n,s),c=this.dom(),l=c.findIndex((t=>t.id.description===e));if(l>=0){const t=this.domItem(l);t&&!t.destroyed&&(t.destroy(),c.splice(l,1))}let h;if("canvas"===a.tagName.toLowerCase())h=a,h.dataset[i]="false";else{const t=a.getElementsByTagName("canvas");t.length?(h=t[0],h.dataset[i]="false"):(h=document.createElement("canvas"),h.dataset[i]="true",a.appendChild(h))}h.style.width||(h.style.width="100%"),h.style.height||(h.style.height="100%");const d=new ki(this,e,r);return l>=0?c.splice(l,0,d):c.push(d),d.canvas.loadCanvas(h),await d.start(),d}loadOptions(t,e){for(const i of this.plugins)i.loadOptions(t,e)}loadParticlesOptions(t,e,...i){const s=this.updaters.get(t);if(s)for(const t of s)t.loadOptions&&t.loadOptions(e,...i)}async refresh(t=!0){t&&this.dom().forEach((t=>t.refresh()))}removeEventListener(t,e){this._eventDispatcher.removeEventListener(t,e)}setOnClickHandler(t){const e=this.dom();if(!e.length)throw new Error(`${f} can only set click handlers after calling tsParticles.load()`);for(const i of e)i.addClickHandler(t)}}class Pi{constructor(){this.key="hsl",this.stringPrefix="hsl"}handleColor(t){const e=t.value.hsl??t.value;if(void 0!==e.h&&void 0!==e.s&&void 0!==e.l)return At(e)}handleRangeColor(t){const e=t.value.hsl??t.value;if(void 0!==e.h&&void 0!==e.l)return At({h:C(e.h),l:C(e.l),s:C(e.s)})}parseString(t){if(!t.startsWith("hsl"))return;const e=/hsla?\(\s*(\d+)\s*,\s*(\d+)%\s*,\s*(\d+)%\s*(,\s*([\d.%]+)\s*)?\)/i.exec(t);return e?Ft({a:e.length>4?H(e[5]):1,h:parseInt(e[1],10),l:parseInt(e[3],10),s:parseInt(e[2],10)}):void 0}}class Oi{constructor(){this.key="rgb",this.stringPrefix="rgb"}handleColor(t){const e=t.value.rgb??t.value;if(void 0!==e.r)return e}handleRangeColor(t){const e=t.value.rgb??t.value;if(void 0!==e.r)return{r:C(e.r),g:C(e.g),b:C(e.b)}}parseString(t){if(!t.startsWith(this.stringPrefix))return;const e=/rgba?\(\s*(\d+)\s*,\s*(\d+)\s*,\s*(\d+)\s*(,\s*([\d.%]+)\s*)?\)/i.exec(t);return e?{a:e.length>4?H(e[5]):1,b:parseInt(e[3],10),g:parseInt(e[2],10),r:parseInt(e[1],10)}:void 0}}class Si{constructor(t){this.container=t,this.type="external"}}class Di{constructor(t){this.container=t,this.type="particles"}}const Ti=function(){const t=new Oi,e=new Pi;Pt(t),Pt(e);const i=new Ci;return i.init(),i}();j()||(window.tsParticles=Ti);class Ri{constructor(){this.radius=0,this.mass=0}load(t){t&&(void 0!==t.mass&&(this.mass=t.mass),void 0!==t.radius&&(this.radius=t.radius))}}class Ei extends Re{constructor(){super(),this.density=5,this.value=50,this.limit=new Ri}load(t){t&&(super.load(t),void 0!==t.density&&(this.density=t.density),wt(t.limit)?this.limit.radius=t.limit:this.limit.load(t.limit))}}class Ii{constructor(){this.color=new ce,this.color.value="#000000",this.draggable=!1,this.opacity=1,this.destroy=!0,this.orbits=!1,this.size=new Ei}load(t){void 0!==t&&(void 0!==t.color&&(this.color=ce.create(this.color,t.color)),void 0!==t.draggable&&(this.draggable=t.draggable),this.name=t.name,void 0!==t.opacity&&(this.opacity=t.opacity),void 0!==t.position&&(this.position={},void 0!==t.position.x&&(this.position.x=S(t.position.x)),void 0!==t.position.y&&(this.position.y=S(t.position.y))),void 0!==t.size&&this.size.load(t.size),void 0!==t.destroy&&(this.destroy=t.destroy),void 0!==t.orbits&&(this.orbits=t.orbits))}}class Li{constructor(t,e,i,s){this.absorbers=t,this.container=e,this._calcPosition=()=>{const t=F({size:this.container.canvas.size,position:this.options.position});return y.create(t.x,t.y)},this._updateParticlePosition=(t,e)=>{if(t.destroyed)return;const i=this.container,s=i.canvas.size;if(t.needsNewPosition){const e=A({size:s});t.position.setTo(e),t.velocity.setTo(t.initialVelocity),t.absorberOrbit=void 0,t.needsNewPosition=!1}if(this.options.orbits){if(void 0===t.absorberOrbit&&(t.absorberOrbit=y.create(0,0),t.absorberOrbit.length=T(t.getPosition(),this.position),t.absorberOrbit.angle=_()*Math.PI*2),t.absorberOrbit.length<=this.size&&!this.options.destroy){const e=Math.min(s.width,s.height);t.absorberOrbit.length=e*(.2*_()-.1+1)}void 0===t.absorberOrbitDirection&&(t.absorberOrbitDirection=t.velocity.x>=0?"clockwise":"counter-clockwise");const o=t.absorberOrbit.length,n=t.absorberOrbit.angle,a=t.absorberOrbitDirection;t.velocity.setTo(y.origin);const r={x:"clockwise"===a?Math.cos:Math.sin,y:"clockwise"===a?Math.sin:Math.cos};t.position.x=this.position.x+o*r.x(n),t.position.y=this.position.y+o*r.y(n),t.absorberOrbit.length-=e.length,t.absorberOrbit.angle+=(t.retina.moveSpeed??0)*i.retina.pixelRatio/100*i.retina.reduceFactor}else{const i=y.origin;i.length=e.length,i.angle=e.angle,t.velocity.addTo(i)}},this.initialPosition=s?y.create(s.x,s.y):void 0,i instanceof Ii?this.options=i:(this.options=new Ii,this.options.load(i)),this.dragging=!1,this.name=this.options.name,this.opacity=this.options.opacity,this.size=C(this.options.size.value)*e.retina.pixelRatio,this.mass=this.size*this.options.size.density*e.retina.reduceFactor;const o=this.options.size.limit;this.limit={radius:o.radius*e.retina.pixelRatio*e.retina.reduceFactor,mass:o.mass},this.color=St(this.options.color)??{b:0,g:0,r:0},this.position=this.initialPosition?.copy()??this._calcPosition()}attract(t){const e=this.container,i=this.options;if(i.draggable){const t=e.interactivity.mouse;if(t.clicking&&t.downPosition){T(this.position,t.downPosition)<=this.size&&(this.dragging=!0)}else this.dragging=!1;this.dragging&&t.position&&(this.position.x=t.position.x,this.position.y=t.position.y)}const s=t.getPosition(),{dx:o,dy:n,distance:a}=D(this.position,s),r=y.create(o,n);if(r.length=this.mass/Math.pow(a,2)*e.retina.reduceFactor,at.getRadius()&&avoid 0===t||wt(t)?this.array[t||0]:this.array.find((e=>e.name===t)),t.addAbsorber=(t,e)=>this.addAbsorber(t,e)}addAbsorber(t,e){const i=new Li(this,this.container,t,e);return this.array.push(i),i}draw(t){for(const e of this.array)e.draw(t)}handleClickMode(t){const e=this.absorbers,i=this.interactivityAbsorbers;if("absorber"===t){const t=ut(i)??ut(e),s=this.container.interactivity.mouse.clickPosition;this.addAbsorber(t,s)}}async init(){this.absorbers=this.container.actualOptions.absorbers,this.interactivityAbsorbers=this.container.actualOptions.interactivity.modes.absorbers,dt(this.absorbers,(t=>{this.addAbsorber(t)}))}particleUpdate(t){for(const e of this.array)if(e.attract(t),t.destroyed)break}removeAbsorber(t){const e=this.array.indexOf(t);e>=0&&this.array.splice(e,1)}resize(){for(const t of this.array)t.resize()}stop(){this.array=[]}}class Fi{constructor(){this.id="absorbers"}getPlugin(t){return new Ai(t)}loadOptions(t,e){(this.needsPlugin(t)||this.needsPlugin(e))&&(e?.absorbers&&(t.absorbers=dt(e.absorbers,(t=>{const e=new Ii;return e.load(t),e}))),t.interactivity.modes.absorbers=dt(e?.interactivity?.modes?.absorbers,(t=>{const e=new Ii;return e.load(t),e})))}needsPlugin(t){if(!t)return!1;const e=t.absorbers;return kt(e)?!!e.length:!!e||!(!t.interactivity?.events?.onClick?.mode||!Z("absorber",t.interactivity.events.onClick.mode))}}class Bi{load(t){t&&(void 0!==t.bottom&&(this.bottom=S(t.bottom)),void 0!==t.left&&(this.left=S(t.left)),void 0!==t.right&&(this.right=S(t.right)),void 0!==t.top&&(this.top=S(t.top)))}}class qi extends Re{constructor(){super(),this.value=3}}class Hi extends Re{constructor(){super(),this.value={min:4,max:9}}}class Vi{constructor(){this.count=1,this.factor=new qi,this.rate=new Hi,this.sizeOffset=!0}load(t){t&&(void 0!==t.color&&(this.color=ce.create(this.color,t.color)),void 0!==t.count&&(this.count=t.count),this.factor.load(t.factor),this.rate.load(t.rate),this.particles=dt(t.particles,(t=>st({},t))),void 0!==t.sizeOffset&&(this.sizeOffset=t.sizeOffset),t.colorOffset&&(this.colorOffset=this.colorOffset??{},void 0!==t.colorOffset.h&&(this.colorOffset.h=t.colorOffset.h),void 0!==t.colorOffset.s&&(this.colorOffset.s=t.colorOffset.s),void 0!==t.colorOffset.l&&(this.colorOffset.l=t.colorOffset.l)))}}class Ui{constructor(){this.bounds=new Bi,this.mode="none",this.split=new Vi}load(t){t&&(t.mode&&(this.mode=t.mode),t.bounds&&this.bounds.load(t.bounds),this.split.load(t.split))}}function Wi(t,e,i,s){const o=i.options.destroy;if(!o)return;const n=o.split,a=ci(t,e,i.options),r=C(n.factor.value),c=i.getFillColor();n.color?a.color.load(n.color):n.colorOffset&&c?a.color.load({value:{hsl:{h:c.h+C(n.colorOffset.h??0),s:c.s+C(n.colorOffset.s??0),l:c.l+C(n.colorOffset.l??0)}}}):a.color.load({value:{hsl:i.getFillColor()}}),a.move.load({center:{x:i.position.x,y:i.position.y,mode:"precise"}}),wt(a.size.value)?a.size.value/=r:(a.size.value.min/=r,a.size.value.max/=r),a.load(s);const l=n.sizeOffset?S(-i.size.value,i.size.value):0,h={x:i.position.x+M(l),y:i.position.y+M(l)};return e.particles.addParticle(h,a,i.group,(t=>!(t.size.value<.5)&&(t.velocity.length=M(S(i.velocity.length,t.velocity.length)),t.splitCount=(i.splitCount??0)+1,t.unbreakable=!0,setTimeout((()=>{t.unbreakable=!1}),500),!0)))}class $i{constructor(t,e){this.engine=t,this.container=e}init(t){const e=this.container,i=t.options.destroy;if(!i)return;t.splitCount=0;const s=i.bounds;t.destroyBounds||(t.destroyBounds={});const{bottom:o,left:n,right:a,top:r}=s,{destroyBounds:c}=t,l=e.canvas.size;o&&(c.bottom=C(o)*l.height/100),n&&(c.left=C(n)*l.width/100),a&&(c.right=C(a)*l.width/100),r&&(c.top=C(r)*l.height/100)}isEnabled(t){return!t.destroyed}loadOptions(t,...e){t.destroy||(t.destroy=new Ui);for(const i of e)t.destroy.load(i?.destroy)}particleDestroyed(t,e){if(e)return;const i=t.options.destroy;i&&"split"===i.mode&&function(t,e,i){const s=i.options.destroy;if(!s)return;const o=s.split;if(o.count>=0&&(void 0===i.splitCount||i.splitCount++>o.count))return;const n=C(o.rate.value),a=ut(o.particles);for(let s=0;s=i.bottom||void 0!==i.left&&e.x<=i.left||void 0!==i.right&&e.x>=i.right||void 0!==i.top&&e.y<=i.top)&&t.destroy()}}class ji{constructor(){this.wait=!1}load(t){t&&(void 0!==t.count&&(this.count=t.count),void 0!==t.delay&&(this.delay=S(t.delay)),void 0!==t.duration&&(this.duration=S(t.duration)),void 0!==t.wait&&(this.wait=t.wait))}}class Gi{constructor(){this.quantity=1,this.delay=.1}load(t){void 0!==t&&(void 0!==t.quantity&&(this.quantity=S(t.quantity)),void 0!==t.delay&&(this.delay=S(t.delay)))}}class Ni{constructor(){this.color=!1,this.opacity=!1}load(t){t&&(void 0!==t.color&&(this.color=t.color),void 0!==t.opacity&&(this.opacity=t.opacity))}}class Xi{constructor(){this.options={},this.replace=new Ni,this.type="square"}load(t){t&&(void 0!==t.options&&(this.options=st({},t.options??{})),this.replace.load(t.replace),void 0!==t.type&&(this.type=t.type))}}class Yi{constructor(){this.mode="percent",this.height=0,this.width=0}load(t){void 0!==t&&(void 0!==t.mode&&(this.mode=t.mode),void 0!==t.height&&(this.height=t.height),void 0!==t.width&&(this.width=t.width))}}class Zi{constructor(){this.autoPlay=!0,this.fill=!0,this.life=new ji,this.rate=new Gi,this.shape=new Xi,this.startCount=0}load(t){t&&(void 0!==t.autoPlay&&(this.autoPlay=t.autoPlay),void 0!==t.size&&(this.size||(this.size=new Yi),this.size.load(t.size)),void 0!==t.direction&&(this.direction=t.direction),this.domId=t.domId,void 0!==t.fill&&(this.fill=t.fill),this.life.load(t.life),this.name=t.name,this.particles=dt(t.particles,(t=>st({},t))),this.rate.load(t.rate),this.shape.load(t.shape),void 0!==t.position&&(this.position={},void 0!==t.position.x&&(this.position.x=S(t.position.x)),void 0!==t.position.y&&(this.position.y=S(t.position.y))),void 0!==t.spawnColor&&(void 0===this.spawnColor&&(this.spawnColor=new Se),this.spawnColor.load(t.spawnColor)),void 0!==t.startCount&&(this.startCount=t.startCount))}}function Qi(t,e){t.color?t.color.value=e:t.color={value:e}}class Ji{constructor(t,e,i,s,o){this.emitters=e,this.container=i,this._destroy=()=>{this._mutationObserver?.disconnect(),this._mutationObserver=void 0,this._resizeObserver?.disconnect(),this._resizeObserver=void 0,this.emitters.removeEmitter(this),this._engine.dispatchEvent("emitterDestroyed",{container:this.container,data:{emitter:this}})},this._prepareToDie=()=>{if(this._paused)return;const t=void 0!==this.options.life?.duration?C(this.options.life.duration):void 0;this.container.retina.reduceFactor&&(this._lifeCount>0||this._immortal)&&void 0!==t&&t>0&&(this._duration=1e3*t)},this._setColorAnimation=(t,e,i)=>{const s=this.container;if(!t.enable)return e;const o=M(t.offset),n=1e3*C(this.options.rate.delay)/s.retina.reduceFactor;return(e+C(t.speed??0)*s.fpsLimit/n+3.6*o)%i},this._engine=t,this._currentDuration=0,this._currentEmitDelay=0,this._currentSpawnDelay=0,this._initialPosition=o,s instanceof Zi?this.options=s:(this.options=new Zi,this.options.load(s)),this._spawnDelay=1e3*C(this.options.life.delay??0)/this.container.retina.reduceFactor,this.position=this._initialPosition??this._calcPosition(),this.name=this.options.name,this.fill=this.options.fill,this._firstSpawn=!this.options.life.wait,this._startParticlesAdded=!1;let n=st({},this.options.particles);if(n??={},n.move??={},n.move.direction??=this.options.direction,this.options.spawnColor&&(this.spawnColor=Rt(this.options.spawnColor)),this._paused=!this.options.autoPlay,this._particlesOptions=n,this._size=this._calcSize(),this.size=mt(this._size,this.container.canvas.size),this._lifeCount=this.options.life.count??-1,this._immortal=this._lifeCount<=0,this.options.domId){const t=document.getElementById(this.options.domId);t&&(this._mutationObserver=new MutationObserver((()=>{this.resize()})),this._resizeObserver=new ResizeObserver((()=>{this.resize()})),this._mutationObserver.observe(t,{attributes:!0,attributeFilter:["style","width","height"]}),this._resizeObserver.observe(t))}const a=this.options.shape,r=this._engine.emitterShapeManager?.getShapeGenerator(a.type);r&&(this._shape=r.generate(this.position,this.size,this.fill,a.options)),this._engine.dispatchEvent("emitterCreated",{container:i,data:{emitter:this}}),this.play()}externalPause(){this._paused=!0,this.pause()}externalPlay(){this._paused=!1,this.play()}async init(){await(this._shape?.init())}pause(){this._paused||delete this._emitDelay}play(){if(!this._paused&&this.container.retina.reduceFactor&&(this._lifeCount>0||this._immortal||!this.options.life.count)&&(this._firstSpawn||this._currentSpawnDelay>=(this._spawnDelay??0))){if(void 0===this._emitDelay){const t=C(this.options.rate.delay);this._emitDelay=1e3*t/this.container.retina.reduceFactor}(this._lifeCount>0||this._immortal)&&this._prepareToDie()}}resize(){const t=this._initialPosition;this.position=t&&tt(t,this.container.canvas.size,y.origin)?t:this._calcPosition(),this._size=this._calcSize(),this.size=mt(this._size,this.container.canvas.size),this._shape?.resize(this.position,this.size)}async update(t){this._paused||(this._firstSpawn&&(this._firstSpawn=!1,this._currentSpawnDelay=this._spawnDelay??0,this._currentEmitDelay=this._emitDelay??0),this._startParticlesAdded||(this._startParticlesAdded=!0,await this._emitParticles(this.options.startCount)),void 0!==this._duration&&(this._currentDuration+=t.value,this._currentDuration>=this._duration&&(this.pause(),void 0!==this._spawnDelay&&delete this._spawnDelay,this._immortal||this._lifeCount--,this._lifeCount>0||this._immortal?(this.position=this._calcPosition(),this._shape?.resize(this.position,this.size),this._spawnDelay=1e3*C(this.options.life.delay??0)/this.container.retina.reduceFactor):this._destroy(),this._currentDuration-=this._duration,delete this._duration)),void 0!==this._spawnDelay&&(this._currentSpawnDelay+=t.value,this._currentSpawnDelay>=this._spawnDelay&&(this._engine.dispatchEvent("emitterPlay",{container:this.container}),this.play(),this._currentSpawnDelay-=this._currentSpawnDelay,delete this._spawnDelay)),void 0!==this._emitDelay&&(this._currentEmitDelay+=t.value,this._currentEmitDelay>=this._emitDelay&&(this._emit(),this._currentEmitDelay-=this._emitDelay)))}_calcPosition(){if(this.options.domId){const t=this.container,e=document.getElementById(this.options.domId);if(e){const i=e.getBoundingClientRect();return{x:(i.x+i.width/2)*t.retina.pixelRatio,y:(i.y+i.height/2)*t.retina.pixelRatio}}}return F({size:this.container.canvas.size,position:this.options.position})}_calcSize(){const t=this.container;if(this.options.domId){const e=document.getElementById(this.options.domId);if(e){const i=e.getBoundingClientRect();return{width:i.width*t.retina.pixelRatio,height:i.height*t.retina.pixelRatio,mode:"precise"}}}return this.options.size??(()=>{const t=new Yi;return t.load({height:0,mode:"percent",width:0}),t})()}async _emit(){if(this._paused)return;const t=C(this.options.rate.quantity);await this._emitParticles(t)}async _emitParticles(t){const e=ut(this._particlesOptions);for(let i=0;ivoid 0===t||wt(t)?this.array[t||0]:this.array.find((e=>e.name===t)),e.addEmitter=async(t,e)=>this.addEmitter(t,e),e.removeEmitter=t=>{const i=e.getEmitter(t);i&&this.removeEmitter(i)},e.playEmitter=t=>{const i=e.getEmitter(t);i&&i.externalPlay()},e.pauseEmitter=t=>{const i=e.getEmitter(t);i&&i.externalPause()}}async addEmitter(t,e){const i=new Zi;i.load(t);const s=new Ji(this._engine,this,this.container,i,e);return await s.init(),this.array.push(s),s}handleClickMode(t){const e=this.emitters,i=this.interactivityEmitters;if("emitter"!==t)return;let s;if(i&&kt(i.value))if(i.value.length>0&&i.random.enable){s=[];const t=[];for(let e=0;e{this.addEmitter(t,n)}))}async init(){if(this.emitters=this.container.actualOptions.emitters,this.interactivityEmitters=this.container.actualOptions.interactivity.modes.emitters,this.emitters)if(kt(this.emitters))for(const t of this.emitters)await this.addEmitter(t);else await this.addEmitter(this.emitters)}pause(){for(const t of this.array)t.pause()}play(){for(const t of this.array)t.play()}removeEmitter(t){const e=this.array.indexOf(t);e>=0&&this.array.splice(e,1)}resize(){for(const t of this.array)t.resize()}stop(){this.array=[]}async update(t){for(const e of this.array)await e.update(t)}}const ts=new Map;class es{constructor(t){this._engine=t}addShapeGenerator(t,e){this.getShapeGenerator(t)||ts.set(t,e)}getShapeGenerator(t){return ts.get(t)}getSupportedShapeGenerators(){return ts.keys()}}class is{constructor(t,e,i,s){this.position=t,this.size=e,this.fill=i,this.options=s}resize(t,e){this.position=t,this.size=e}}class ss{constructor(t){this._engine=t,this.id="emitters"}getPlugin(t){return new Ki(this._engine,t)}loadOptions(t,e){if(!this.needsPlugin(t)&&!this.needsPlugin(e))return;e?.emitters&&(t.emitters=dt(e.emitters,(t=>{const e=new Zi;return e.load(t),e})));const i=e?.interactivity?.modes?.emitters;if(i)if(kt(i))t.interactivity.modes.emitters={random:{count:1,enable:!0},value:i.map((t=>{const e=new Zi;return e.load(t),e}))};else{const e=i;if(void 0!==e.value)if(kt(e.value))t.interactivity.modes.emitters={random:{count:e.random.count??1,enable:e.random.enable??!1},value:e.value.map((t=>{const e=new Zi;return e.load(t),e}))};else{const i=new Zi;i.load(e.value),t.interactivity.modes.emitters={random:{count:e.random.count??1,enable:e.random.enable??!1},value:i}}else{(t.interactivity.modes.emitters={random:{count:1,enable:!1},value:new Zi}).value.load(i)}}}needsPlugin(t){if(!t)return!1;const e=t.emitters;return kt(e)&&!!e.length||void 0!==e||!!t.interactivity?.events?.onClick?.mode&&Z("emitter",t.interactivity.events.onClick.mode)}}class os extends is{constructor(t,e,i,s){super(t,e,i,s)}async init(){}async randomPosition(){const t=this.size,e=this.fill,i=this.position,[s,o]=[t.width/2,t.height/2],n=((t,e)=>{const i=_()/4,s=Math.atan(e/t*Math.tan(2*Math.PI*i)),o=_();return o<.25?s:o<.5?Math.PI-s:o<.75?Math.PI+s:-s})(s,o),a=(h=n,(c=s)*(l=o)/Math.sqrt((l*Math.cos(h))**2+(c*Math.sin(h))**2)),r=e?a*Math.sqrt(_()):a;var c,l,h;return{position:{x:i.x+r*Math.cos(n),y:i.y+r*Math.sin(n)}}}}class ns{generate(t,e,i,s){return new os(t,e,i,s)}}function as(t,e){return t+e*(_()-.5)}class rs extends is{constructor(t,e,i,s){super(t,e,i,s)}async init(){}async randomPosition(){const t=this.fill,e=this.position,i=this.size;if(t)return{position:{x:as(e.x,i.width),y:as(e.y,i.height)}};{const t=i.width/2,s=i.height/2,o=Math.floor(4*_()),n=2*(_()-.5);switch(o){case 0:return{position:{x:e.x+n*t,y:e.y-s}};case 1:return{position:{x:e.x-t,y:e.y+n*s}};case 2:return{position:{x:e.x+n*t,y:e.y+s}};default:return{position:{x:e.x+t,y:e.y+n*s}}}}}}class cs{generate(t,e,i,s){return new rs(t,e,i,s)}}class ls{constructor(){this.delay=1,this.pauseOnStop=!1,this.quantity=1}load(t){t&&(void 0!==t.delay&&(this.delay=t.delay),void 0!==t.quantity&&(this.quantity=t.quantity),void 0!==t.particles&&(this.particles=st({},t.particles)),void 0!==t.pauseOnStop&&(this.pauseOnStop=t.pauseOnStop))}}const hs="trail";class ds extends Si{constructor(t){super(t),this._delay=0}clear(){}init(){}async interact(t){const e=this.container,{interactivity:i}=e;if(!e.retina.reduceFactor)return;const s=e.actualOptions.interactivity.modes.trail;if(!s)return;const o=1e3*s.delay/this.container.retina.reduceFactor;if(this._delay=.5?"darken":"enlighten";t.roll.alter={type:i,value:C("darken"===i?e.darken.value:e.enlighten.value)}}else e.darken.enable?t.roll.alter={type:"darken",value:C(e.darken.value)}:e.enlighten.enable&&(t.roll.alter={type:"enlighten",value:C(e.enlighten.value)});else t.roll={enable:!1,horizontal:!1,vertical:!1,angle:0,speed:0}}(t)}isEnabled(t){const e=t.options.roll;return!t.destroyed&&!t.spawning&&!!e?.enable}loadOptions(t,...e){t.roll||(t.roll=new ps);for(const i of e)t.roll.load(i?.roll)}update(t,e){this.isEnabled(t)&&function(t,e){const i=t.options.roll,s=t.roll;if(!s||!i?.enable)return;const o=s.speed*e.factor,n=2*Math.PI;s.angle+=o,s.angle>n&&(s.angle-=n)}(t,e)}}function vs(t,e,i,s,o,n){!function(t,e){const i=t.options,s=i.move.path;if(!s.enable)return;if(t.lastPathTime<=t.pathDelay)return void(t.lastPathTime+=e.value);const o=t.pathGenerator?.generate(t,e);o&&t.velocity.addTo(o);s.clamp&&(t.velocity.x=k(t.velocity.x,-1,1),t.velocity.y=k(t.velocity.y,-1,1));t.lastPathTime-=t.pathDelay}(t,n);const a=t.gravity,r=a?.enable&&a.inverse?-1:1;o&&i&&(t.velocity.x+=o*n.factor/(60*i)),a?.enable&&i&&(t.velocity.y+=r*(a.acceleration*n.factor)/(60*i));const c=t.moveDecay;t.velocity.multTo(c);const l=t.velocity.mult(i);a?.enable&&s>0&&(!a.inverse&&l.y>=0&&l.y>=s||a.inverse&&l.y<=0&&l.y<=-s)&&(l.y=r*s,i&&(t.velocity.y=l.y/i));const h=t.options.zIndex,d=(1-t.zIndexFactor)**h.velocityRate;l.multTo(d);const{position:u}=t;u.addTo(l),e.vibrate&&(u.x+=Math.sin(u.x*Math.cos(u.y)),u.y+=Math.cos(u.y*Math.sin(u.x)))}class ys{constructor(){this._initSpin=t=>{const e=t.container,i=t.options.move.spin;if(!i.enable)return;const s=i.position??{x:50,y:50},o={x:.01*s.x*e.canvas.size.width,y:.01*s.y*e.canvas.size.height},n=T(t.getPosition(),o),a=C(i.acceleration);t.retina.spinAcceleration=a*e.retina.pixelRatio,t.spin={center:o,direction:t.velocity.x>=0?"clockwise":"counter-clockwise",angle:t.velocity.angle,radius:n,acceleration:t.retina.spinAcceleration}}}init(t){const e=t.options.move.gravity;t.gravity={enable:e.enable,acceleration:C(e.acceleration),inverse:e.inverse},this._initSpin(t)}isEnabled(t){return!t.destroyed&&t.options.move.enable}move(t,e){const i=t.options,s=i.move;if(!s.enable)return;const o=t.container,n=o.retina.pixelRatio,a=function(t){return t.slow.inRange?t.slow.factor:1}(t),r=(t.retina.moveSpeed??=C(s.speed)*n)*o.retina.reduceFactor,c=t.retina.moveDrift??=C(t.options.move.drift)*n,l=O(i.size.value)*n,h=r*(s.size?t.getRadius()/l:1)*a*(e.factor||1)/2,d=t.retina.maxSpeed??o.retina.maxSpeed;s.spin.enable?function(t,e){const i=t.container;if(!t.spin)return;const s={x:"clockwise"===t.spin.direction?Math.cos:Math.sin,y:"clockwise"===t.spin.direction?Math.sin:Math.cos};t.position.x=t.spin.center.x+t.spin.radius*s.x(t.spin.angle),t.position.y=t.spin.center.y+t.spin.radius*s.y(t.spin.angle),t.spin.radius+=t.spin.acceleration;const o=Math.max(i.canvas.size.width,i.canvas.size.height),n=.5*o;t.spin.radius>n?(t.spin.radius=n,t.spin.acceleration*=-1):t.spin.radius<0&&(t.spin.radius=0,t.spin.acceleration*=-1),t.spin.angle+=.01*e*(1-t.spin.radius/o)}(t,h):vs(t,s,h,d,c,e),function(t){const e=t.initialPosition,{dx:i,dy:s}=D(e,t.position),o=Math.abs(i),n=Math.abs(s),{maxDistance:a}=t.retina,r=a.horizontal,c=a.vertical;if(r||c)if((r&&o>=r||c&&n>=c)&&!t.misplaced)t.misplaced=!!r&&o>r||!!c&&n>c,r&&(t.velocity.x=.5*t.velocity.y-t.velocity.x),c&&(t.velocity.y=.5*t.velocity.x-t.velocity.y);else if((!r||oe.x&&s.x>0)&&(s.x*=-_()),c&&(i.ye.y&&s.y>0)&&(s.y*=-_())}}(t)}}class ms{draw(t){const{context:e,particle:i,radius:s}=t;i.circleRange||(i.circleRange={min:0,max:2*Math.PI});const o=i.circleRange;e.arc(0,0,s,o.min,o.max,!1)}getSidesCount(){return 12}particleInit(t,e){const i=e.shapeData,s=i?.angle??{max:360,min:0};e.circleRange=_t(s)?{min:s.min*Math.PI/180,max:s.max*Math.PI/180}:{min:0,max:s*Math.PI/180}}}function gs(t,e,i,s,o){if(!e||!i.enable||(e.maxLoops??0)>0&&(e.loops??0)>(e.maxLoops??0))return;if(e.time||(e.time=0),(e.delayTime??0)>0&&e.time<(e.delayTime??0)&&(e.time+=t.value),(e.delayTime??0)>0&&e.time<(e.delayTime??0))return;const n=M(i.offset),a=(e.velocity??0)*t.factor+3.6*n,r=e.decay??1;o&&"increasing"!==e.status?(e.value-=a,e.value<0&&(e.loops||(e.loops=0),e.loops++,e.status="increasing",e.value+=e.value)):(e.value+=a,e.value>s&&(e.loops||(e.loops=0),e.loops++,o&&(e.status="decreasing",e.value-=e.value%s))),e.velocity&&1!==r&&(e.velocity*=r),e.value>s&&(e.value%=s)}class bs{constructor(t){this.container=t}init(t){const e=Rt(t.options.color,t.id,t.options.reduceDuplicates);e&&(t.color=jt(e,t.options.color.animation,this.container.retina.reduceFactor))}isEnabled(t){const{h:e,s:i,l:s}=t.options.color.animation,{color:o}=t;return!t.destroyed&&!t.spawning&&(void 0!==o?.h.value&&e.enable||void 0!==o?.s.value&&i.enable||void 0!==o?.l.value&&s.enable)}update(t,e){!function(t,e){const{h:i,s,l:o}=t.options.color.animation,{color:n}=t;if(!n)return;const{h:a,s:r,l:c}=n;a&&gs(e,a,i,360,!1),r&&gs(e,r,s,100,!0),c&&gs(e,c,o,100,!0)}(t,e)}}class ws{constructor(t){this.container=t}init(t){const e=t.options.opacity;t.opacity=ft(e,1);const i=e.animation;i.enable&&(t.opacity.velocity=C(i.speed)/100*this.container.retina.reduceFactor,i.sync||(t.opacity.velocity*=_()))}isEnabled(t){return!t.destroyed&&!t.spawning&&!!t.opacity&&t.opacity.enable&&((t.opacity.maxLoops??0)<=0||(t.opacity.maxLoops??0)>0&&(t.opacity.loops??0)<(t.opacity.maxLoops??0))}reset(t){t.opacity&&(t.opacity.time=0,t.opacity.loops=0)}update(t,e){this.isEnabled(t)&&function(t,e){const i=t.opacity;if(t.destroyed||!i?.enable||(i.maxLoops??0)>0&&(i.loops??0)>(i.maxLoops??0))return;const s=i.min,o=i.max,n=i.decay??1;if(i.time||(i.time=0),(i.delayTime??0)>0&&i.time<(i.delayTime??0)&&(i.time+=e.value),!((i.delayTime??0)>0&&i.time<(i.delayTime??0))){switch(i.status){case"increasing":i.value>=o?(i.status="decreasing",i.loops||(i.loops=0),i.loops++):i.value+=(i.velocity??0)*e.factor;break;case"decreasing":i.value<=s?(i.status="increasing",i.loops||(i.loops=0),i.loops++):i.value-=(i.velocity??0)*e.factor}i.velocity&&1!==i.decay&&(i.velocity*=n),function(t,e,i,s){switch(t.options.opacity.animation.destroy){case"max":e>=s&&t.destroy();break;case"min":e<=i&&t.destroy()}}(t,i.value,s,o),t.destroyed||(i.value=k(i.value,s,o))}}(t,e)}}class xs{constructor(t){this.container=t,this.modes=["bounce","bounce-vertical","bounce-horizontal","bounceVertical","bounceHorizontal","split"]}update(t,e,i,s){if(!this.modes.includes(s))return;const o=this.container;let n=!1;for(const[,s]of o.plugins)if(void 0!==s.particleBounce&&(n=s.particleBounce(t,i,e)),n)break;if(n)return;const a=t.getPosition(),r=t.offset,c=t.getRadius(),l=it(a,c),h=o.canvas.size;!function(t){if("bounce"!==t.outMode&&"bounce-horizontal"!==t.outMode&&"bounceHorizontal"!==t.outMode&&"split"!==t.outMode||"left"!==t.direction&&"right"!==t.direction)return;t.bounds.right<0&&"left"===t.direction?t.particle.position.x=t.size+t.offset.x:t.bounds.left>t.canvasSize.width&&"right"===t.direction&&(t.particle.position.x=t.canvasSize.width-t.size-t.offset.x);const e=t.particle.velocity.x;let i=!1;if("right"===t.direction&&t.bounds.right>=t.canvasSize.width&&e>0||"left"===t.direction&&t.bounds.left<=0&&e<0){const e=C(t.particle.options.bounce.horizontal.value);t.particle.velocity.x*=-e,i=!0}if(!i)return;const s=t.offset.x+t.size;t.bounds.right>=t.canvasSize.width&&"right"===t.direction?t.particle.position.x=t.canvasSize.width-s:t.bounds.left<=0&&"left"===t.direction&&(t.particle.position.x=s),"split"===t.outMode&&t.particle.destroy()}({particle:t,outMode:s,direction:e,bounds:l,canvasSize:h,offset:r,size:c}),function(t){if("bounce"!==t.outMode&&"bounce-vertical"!==t.outMode&&"bounceVertical"!==t.outMode&&"split"!==t.outMode||"bottom"!==t.direction&&"top"!==t.direction)return;t.bounds.bottom<0&&"top"===t.direction?t.particle.position.y=t.size+t.offset.y:t.bounds.top>t.canvasSize.height&&"bottom"===t.direction&&(t.particle.position.y=t.canvasSize.height-t.size-t.offset.y);const e=t.particle.velocity.y;let i=!1;if("bottom"===t.direction&&t.bounds.bottom>=t.canvasSize.height&&e>0||"top"===t.direction&&t.bounds.top<=0&&e<0){const e=C(t.particle.options.bounce.vertical.value);t.particle.velocity.y*=-e,i=!0}if(!i)return;const s=t.offset.y+t.size;t.bounds.bottom>=t.canvasSize.height&&"bottom"===t.direction?t.particle.position.y=t.canvasSize.height-s:t.bounds.top<=0&&"top"===t.direction&&(t.particle.position.y=s),"split"===t.outMode&&t.particle.destroy()}({particle:t,outMode:s,direction:e,bounds:l,canvasSize:h,offset:r,size:c})}}class _s{constructor(t){this.container=t,this.modes=["destroy"]}update(t,e,i,s){if(!this.modes.includes(s))return;const o=this.container;switch(t.outType){case"normal":case"outside":if(tt(t.position,o.canvas.size,y.origin,t.getRadius(),e))return;break;case"inside":{const{dx:e,dy:i}=D(t.position,t.moveCenter),{x:s,y:o}=t.velocity;if(s<0&&e>t.moveCenter.radius||o<0&&i>t.moveCenter.radius||s>=0&&e<-t.moveCenter.radius||o>=0&&i<-t.moveCenter.radius)return;break}}o.particles.remove(t,void 0,!0)}}class ks{constructor(t){this.container=t,this.modes=["none"]}update(t,e,i,s){if(!this.modes.includes(s))return;if(t.options.move.distance.horizontal&&("left"===e||"right"===e)||t.options.move.distance.vertical&&("top"===e||"bottom"===e))return;const o=t.options.move.gravity,n=this.container,a=n.canvas.size,r=t.getRadius();if(o.enable){const i=t.position;(!o.inverse&&i.y>a.height+r&&"bottom"===e||o.inverse&&i.y<-r&&"top"===e)&&n.particles.remove(t)}else{if(t.velocity.y>0&&t.position.y<=a.height+r||t.velocity.y<0&&t.position.y>=-r||t.velocity.x>0&&t.position.x<=a.width+r||t.velocity.x<0&&t.position.x>=-r)return;tt(t.position,n.canvas.size,y.origin,r,e)||n.particles.remove(t)}}}class zs{constructor(t){this.container=t,this.modes=["out"]}update(t,e,i,s){if(!this.modes.includes(s))return;const o=this.container;switch(t.outType){case"inside":{const{x:e,y:i}=t.velocity,s=y.origin;s.length=t.moveCenter.radius,s.angle=t.velocity.angle+Math.PI,s.addTo(y.create(t.moveCenter));const{dx:n,dy:a}=D(t.position,s);if(e<=0&&n>=0||i<=0&&a>=0||e>=0&&n<=0||i>=0&&a<=0)return;t.position.x=Math.floor(M({min:0,max:o.canvas.size.width})),t.position.y=Math.floor(M({min:0,max:o.canvas.size.height}));const{dx:r,dy:c}=D(t.position,t.moveCenter);t.direction=Math.atan2(-c,-r),t.velocity.angle=t.direction;break}default:if(tt(t.position,o.canvas.size,y.origin,t.getRadius(),e))return;switch(t.outType){case"outside":{t.position.x=Math.floor(M({min:-t.moveCenter.radius,max:t.moveCenter.radius}))+t.moveCenter.x,t.position.y=Math.floor(M({min:-t.moveCenter.radius,max:t.moveCenter.radius}))+t.moveCenter.y;const{dx:e,dy:i}=D(t.position,t.moveCenter);t.moveCenter.radius&&(t.direction=Math.atan2(i,e),t.velocity.angle=t.direction);break}case"normal":{const i=t.options.move.warp,s=o.canvas.size,n={bottom:s.height+t.getRadius()+t.offset.y,left:-t.getRadius()-t.offset.x,right:s.width+t.getRadius()+t.offset.x,top:-t.getRadius()-t.offset.y},a=t.getRadius(),r=it(t.position,a);"right"===e&&r.left>s.width+t.offset.x?(t.position.x=n.left,t.initialPosition.x=t.position.x,i||(t.position.y=_()*s.height,t.initialPosition.y=t.position.y)):"left"===e&&r.right<-t.offset.x&&(t.position.x=n.right,t.initialPosition.x=t.position.x,i||(t.position.y=_()*s.height,t.initialPosition.y=t.position.y)),"bottom"===e&&r.top>s.height+t.offset.y?(i||(t.position.x=_()*s.width,t.initialPosition.x=t.position.x),t.position.y=n.top,t.initialPosition.y=t.position.y):"top"===e&&r.bottom<-t.offset.y&&(i||(t.position.x=_()*s.width,t.initialPosition.x=t.position.x),t.position.y=n.bottom,t.initialPosition.y=t.position.y);break}}}}}class Ms{constructor(t){this.container=t,this._updateOutMode=(t,e,i,s)=>{for(const o of this.updaters)o.update(t,s,e,i)},this.updaters=[new xs(t),new _s(t),new zs(t),new ks(t)]}init(){}isEnabled(t){return!t.destroyed&&!t.spawning}update(t,e){const i=t.options.move.outModes;this._updateOutMode(t,e,i.bottom??i.default,"bottom"),this._updateOutMode(t,e,i.left??i.default,"left"),this._updateOutMode(t,e,i.right??i.default,"right"),this._updateOutMode(t,e,i.top??i.default,"top")}}class Cs{init(t){const e=t.container,i=t.options.size.animation;i.enable&&(t.size.velocity=(t.retina.sizeAnimationSpeed??e.retina.sizeAnimationSpeed)/100*e.retina.reduceFactor,i.sync||(t.size.velocity*=_()))}isEnabled(t){return!t.destroyed&&!t.spawning&&t.size.enable&&((t.size.maxLoops??0)<=0||(t.size.maxLoops??0)>0&&(t.size.loops??0)<(t.size.maxLoops??0))}reset(t){t.size.loops=0}update(t,e){this.isEnabled(t)&&function(t,e){const i=t.size;if(t.destroyed||!i||!i.enable||(i.maxLoops??0)>0&&(i.loops??0)>(i.maxLoops??0))return;const s=(i.velocity??0)*e.factor,o=i.min,n=i.max,a=i.decay??1;if(i.time||(i.time=0),(i.delayTime??0)>0&&i.time<(i.delayTime??0)&&(i.time+=e.value),!((i.delayTime??0)>0&&i.time<(i.delayTime??0))){switch(i.status){case"increasing":i.value>=n?(i.status="decreasing",i.loops||(i.loops=0),i.loops++):i.value+=s;break;case"decreasing":i.value<=o?(i.status="increasing",i.loops||(i.loops=0),i.loops++):i.value-=s}i.velocity&&1!==a&&(i.velocity*=a),function(t,e,i,s){switch(t.options.size.animation.destroy){case"max":e>=s&&t.destroy();break;case"min":e<=i&&t.destroy()}}(t,i.value,o,n),t.destroyed||(i.value=k(i.value,o,n))}}(t,e)}}async function Ps(t,e=!0){await async function(t,e=!0){await t.addMover("base",(()=>new ys),e)}(t,!1),await async function(t,e=!0){await t.addShape("circle",new ms,e)}(t,!1),await async function(t,e=!0){await t.addParticleUpdater("color",(t=>new bs(t)),e)}(t,!1),await async function(t,e=!0){await t.addParticleUpdater("opacity",(t=>new ws(t)),e)}(t,!1),await async function(t,e=!0){await t.addParticleUpdater("outModes",(t=>new Ms(t)),e)}(t,!1),await async function(t,e=!0){await t.addParticleUpdater("size",(()=>new Cs),e)}(t,!1),await t.refresh(e)}const Os=["emoji"],Ss='"Twemoji Mozilla", Apple Color Emoji, "Segoe UI Emoji", "Noto Color Emoji", "EmojiOne Color"';class Ds{constructor(){this._emojiShapeDict=new Map}destroy(){for(const[,t]of this._emojiShapeDict)t instanceof ImageBitmap&&t?.close()}draw(t){const{context:e,particle:i,radius:s,opacity:o}=t,n=i.emojiData;n&&(e.globalAlpha=o,e.drawImage(n,-s,-s,2*s,2*s),e.globalAlpha=1)}async init(t){const e=t.actualOptions;if(Os.find((t=>Z(t,e.particles.shape.type)))){const t=[Q(Ss)],i=Os.map((t=>e.particles.shape.options[t])).find((t=>!!t));i&&dt(i,(e=>{e.font&&t.push(Q(e.font))})),await Promise.all(t)}}particleDestroy(t){delete t.emojiData}particleInit(t,e){if(!e.emojiData){const t=e.shapeData;if(!t?.value)return;const i=ut(t.value,e.randomIndexData),s=t.font??Ss;if(!i)return;const o=`${i}_${s}`,n=this._emojiShapeDict.get(o);if(n)return void(e.emojiData=n);const a=2*O(e.size.value);let r;if("undefined"!=typeof OffscreenCanvas){const t=new OffscreenCanvas(a,a),o=t.getContext("2d");if(!o)return;o.font=`400 ${2*O(e.size.value)}px ${s}`,o.textBaseline="middle",o.textAlign="center",o.fillText(i,O(e.size.value),O(e.size.value)),r=t.transferToImageBitmap()}else{const t=document.createElement("canvas");t.width=a,t.height=a;const o=t.getContext("2d");if(!o)return;o.font=`400 ${2*O(e.size.value)}px ${s}`,o.textBaseline="middle",o.textAlign="center",o.fillText(i,O(e.size.value),O(e.size.value)),r=t}this._emojiShapeDict.set(o,r),e.emojiData=r}}}class Ts{constructor(){this.distance=200,this.duration=.4,this.easing="ease-out-quad",this.factor=1,this.maxSpeed=50,this.speed=1}load(t){t&&(void 0!==t.distance&&(this.distance=t.distance),void 0!==t.duration&&(this.duration=t.duration),void 0!==t.easing&&(this.easing=t.easing),void 0!==t.factor&&(this.factor=t.factor),void 0!==t.maxSpeed&&(this.maxSpeed=t.maxSpeed),void 0!==t.speed&&(this.speed=t.speed))}}const Rs="attract";class Es extends Si{constructor(t,e){super(e),this._clickAttract=()=>{const t=this.container;t.attract||(t.attract={particles:[]});const{attract:e}=t;if(e.finish||(e.count||(e.count=0),e.count++,e.count===t.particles.count&&(e.finish=!0)),e.clicking){const e=t.interactivity.mouse.clickPosition,i=t.retina.attractModeDistance;if(!i||i<0||!e)return;this._processAttract(e,i,new yi(e.x,e.y,i))}else!1===e.clicking&&(e.particles=[])},this._hoverAttract=()=>{const t=this.container,e=t.interactivity.mouse.position,i=t.retina.attractModeDistance;!i||i<0||!e||this._processAttract(e,i,new yi(e.x,e.y,i))},this._processAttract=(t,e,i)=>{const s=this.container,o=s.actualOptions.interactivity.modes.attract;if(!o)return;const n=s.particles.quadTree.query(i,(t=>this.isEnabled(t)));for(const i of n){const{dx:s,dy:n,distance:a}=D(i.position,t),r=o.speed*o.factor,c=k(w(o.easing)(1-a/e)*r,0,o.maxSpeed),l=y.create(0===a?r:s/a*c,0===a?r:n/a*c);i.position.subFrom(l)}},this._engine=t,e.attract||(e.attract={particles:[]}),this.handleClickMode=t=>{const i=this.container.actualOptions.interactivity.modes.attract;if(i&&t===Rs){e.attract||(e.attract={particles:[]}),e.attract.clicking=!0,e.attract.count=0;for(const t of e.attract.particles)this.isEnabled(t)&&t.velocity.setTo(t.initialVelocity);e.attract.particles=[],e.attract.finish=!1,setTimeout((()=>{e.destroyed||(e.attract||(e.attract={particles:[]}),e.attract.clicking=!1)}),1e3*i.duration)}}}clear(){}init(){const t=this.container,e=t.actualOptions.interactivity.modes.attract;e&&(t.retina.attractModeDistance=e.distance*t.retina.pixelRatio)}async interact(){const t=this.container,e=t.actualOptions,i=t.interactivity.status===r,s=e.interactivity.events,o=s.onHover.enable,n=s.onHover.mode,a=s.onClick.enable,c=s.onClick.mode;i&&o&&Z(Rs,n)?this._hoverAttract():a&&Z(Rs,c)&&this._clickAttract()}isEnabled(t){const e=this.container,i=e.actualOptions,s=e.interactivity.mouse,o=(t?.interactivity??i.interactivity).events;if(!(s.position&&o.onHover.enable||s.clickPosition&&o.onClick.enable))return!1;const n=o.onHover.mode,a=o.onClick.mode;return Z(Rs,n)||Z(Rs,a)}loadModeOptions(t,...e){t.attract||(t.attract=new Ts);for(const i of e)t.attract.load(i?.attract)}reset(){}}class Is{constructor(){this.distance=200}load(t){t&&void 0!==t.distance&&(this.distance=t.distance)}}const Ls="bounce";class As extends Si{constructor(t){super(t),this._processBounce=(t,e,i)=>{const s=this.container.particles.quadTree.query(i,(t=>this.isEnabled(t)));for(const o of s)i instanceof yi?lt(ct(o),{position:t,radius:e,mass:e**2*Math.PI/2,velocity:y.origin,factor:y.origin}):i instanceof vi&&ht(o,it(t,e))},this._processMouseBounce=()=>{const t=this.container,e=10*t.retina.pixelRatio,i=t.interactivity.mouse.position,s=t.retina.bounceModeDistance;!s||s<0||!i||this._processBounce(i,s,new yi(i.x,i.y,s+e))},this._singleSelectorBounce=(t,e)=>{const i=this.container,s=document.querySelectorAll(t);s.length&&s.forEach((t=>{const s=t,o=i.retina.pixelRatio,n={x:(s.offsetLeft+s.offsetWidth/2)*o,y:(s.offsetTop+s.offsetHeight/2)*o},a=s.offsetWidth/2*o,r=10*o,c="circle"===e.type?new yi(n.x,n.y,a+r):new vi(s.offsetLeft*o-r,s.offsetTop*o-r,s.offsetWidth*o+2*r,s.offsetHeight*o+2*r);this._processBounce(n,a,c)}))}}clear(){}init(){const t=this.container,e=t.actualOptions.interactivity.modes.bounce;e&&(t.retina.bounceModeDistance=e.distance*t.retina.pixelRatio)}async interact(){const t=this.container,e=t.actualOptions.interactivity.events,i=t.interactivity.status===r,s=e.onHover.enable,o=e.onHover.mode,n=e.onDiv;i&&s&&Z(Ls,o)?this._processMouseBounce():nt(Ls,n,((t,e)=>this._singleSelectorBounce(t,e)))}isEnabled(t){const e=this.container,i=e.actualOptions,s=e.interactivity.mouse,o=(t?.interactivity??i.interactivity).events,n=o.onDiv;return s.position&&o.onHover.enable&&Z(Ls,o.onHover.mode)||ot(Ls,n)}loadModeOptions(t,...e){t.bounce||(t.bounce=new Is);for(const i of e)t.bounce.load(i?.bounce)}reset(){}}class Fs{constructor(){this.distance=200,this.duration=.4,this.mix=!1}load(t){if(t){if(void 0!==t.distance&&(this.distance=t.distance),void 0!==t.duration&&(this.duration=t.duration),void 0!==t.mix&&(this.mix=t.mix),void 0!==t.opacity&&(this.opacity=t.opacity),void 0!==t.color){const e=kt(this.color)?void 0:this.color;this.color=dt(t.color,(t=>ce.create(e,t)))}void 0!==t.size&&(this.size=t.size)}}}class Bs extends Fs{constructor(){super(),this.selectors=[]}load(t){super.load(t),t&&void 0!==t.selectors&&(this.selectors=t.selectors)}}class qs extends Fs{load(t){super.load(t),t&&(this.divs=dt(t.divs,(t=>{const e=new Bs;return e.load(t),e})))}}function Hs(t,e,i,s){if(e>=i){return k(t+(e-i)*s,t,e)}if(e{const t=this.container,e=t.actualOptions,i=t.interactivity.mouse.clickPosition,s=e.interactivity.modes.bubble;if(!s||!i)return;t.bubble||(t.bubble={});const o=t.retina.bubbleModeDistance;if(!o||o<0)return;const n=t.particles.quadTree.queryCircle(i,o,(t=>this.isEnabled(t))),{bubble:a}=t;for(const e of n){if(!a.clicking)continue;e.bubble.inRange=!a.durationEnd;const n=T(e.getPosition(),i),r=((new Date).getTime()-(t.interactivity.mouse.clickTime||0))/1e3;r>s.duration&&(a.durationEnd=!0),r>2*s.duration&&(a.clicking=!1,a.durationEnd=!1);const c={bubbleObj:{optValue:t.retina.bubbleModeSize,value:e.bubble.radius},particlesObj:{optValue:O(e.options.size.value)*t.retina.pixelRatio,value:e.size.value},type:"size"};this._process(e,n,r,c);const l={bubbleObj:{optValue:s.opacity,value:e.bubble.opacity},particlesObj:{optValue:O(e.options.opacity.value),value:e.opacity?.value??1},type:"opacity"};this._process(e,n,r,l),!a.durationEnd&&n<=o?this._hoverBubbleColor(e,n):delete e.bubble.color}},this._hoverBubble=()=>{const t=this.container,e=t.interactivity.mouse.position,i=t.retina.bubbleModeDistance;if(!i||i<0||void 0===e)return;const s=t.particles.quadTree.queryCircle(e,i,(t=>this.isEnabled(t)));for(const o of s){o.bubble.inRange=!0;const s=T(o.getPosition(),e),a=1-s/i;s<=i?a>=0&&t.interactivity.status===r&&(this._hoverBubbleSize(o,a),this._hoverBubbleOpacity(o,a),this._hoverBubbleColor(o,a)):this.reset(o),t.interactivity.status===n&&this.reset(o)}},this._hoverBubbleColor=(t,e,i)=>{const s=this.container.actualOptions,o=i??s.interactivity.modes.bubble;if(o){if(!t.bubble.finalColor){const e=o.color;if(!e)return;const i=ut(e);t.bubble.finalColor=Rt(i)}if(t.bubble.finalColor)if(o.mix){t.bubble.color=void 0;const i=t.getFillColor();t.bubble.color=i?Et(Vt(i,t.bubble.finalColor,1-e,e)):t.bubble.finalColor}else t.bubble.color=t.bubble.finalColor}},this._hoverBubbleOpacity=(t,e,i)=>{const s=this.container.actualOptions,o=i?.opacity??s.interactivity.modes.bubble?.opacity;if(!o)return;const n=t.options.opacity.value,a=Hs(t.opacity?.value??1,o,O(n),e);void 0!==a&&(t.bubble.opacity=a)},this._hoverBubbleSize=(t,e,i)=>{const s=this.container,o=i?.size?i.size*s.retina.pixelRatio:s.retina.bubbleModeSize;if(void 0===o)return;const n=O(t.options.size.value)*s.retina.pixelRatio,a=Hs(t.size.value,o,n,e);void 0!==a&&(t.bubble.radius=a)},this._process=(t,e,i,s)=>{const o=this.container,n=s.bubbleObj.optValue,a=o.actualOptions.interactivity.modes.bubble;if(!a||void 0===n)return;const r=a.duration,c=o.retina.bubbleModeDistance,l=s.particlesObj.optValue,h=s.bubbleObj.value,d=s.particlesObj.value||0,u=s.type;if(c&&!(c<0)&&n!==l)if(o.bubble||(o.bubble={}),o.bubble.durationEnd)h&&("size"===u&&delete t.bubble.radius,"opacity"===u&&delete t.bubble.opacity);else if(e<=c){if((h??d)!==n){const e=d-i*(d-n)/r;"size"===u&&(t.bubble.radius=e),"opacity"===u&&(t.bubble.opacity=e)}}else"size"===u&&delete t.bubble.radius,"opacity"===u&&delete t.bubble.opacity},this._singleSelectorHover=(t,e,i)=>{const s=this.container,o=document.querySelectorAll(e),n=s.actualOptions.interactivity.modes.bubble;n&&o.length&&o.forEach((e=>{const o=e,a=s.retina.pixelRatio,r={x:(o.offsetLeft+o.offsetWidth/2)*a,y:(o.offsetTop+o.offsetHeight/2)*a},c=o.offsetWidth/2*a,l="circle"===i.type?new yi(r.x,r.y,c):new vi(o.offsetLeft*a,o.offsetTop*a,o.offsetWidth*a,o.offsetHeight*a),h=s.particles.quadTree.query(l,(t=>this.isEnabled(t)));for(const e of h){if(!l.contains(e.getPosition()))continue;e.bubble.inRange=!0;const i=rt(n.divs,o);e.bubble.div&&e.bubble.div===o||(this.clear(e,t,!0),e.bubble.div=o),this._hoverBubbleSize(e,1,i),this._hoverBubbleOpacity(e,1,i),this._hoverBubbleColor(e,1,i)}}))},t.bubble||(t.bubble={}),this.handleClickMode=e=>{e===Vs&&(t.bubble||(t.bubble={}),t.bubble.clicking=!0)}}clear(t,e,i){t.bubble.inRange&&!i||(delete t.bubble.div,delete t.bubble.opacity,delete t.bubble.radius,delete t.bubble.color)}init(){const t=this.container,e=t.actualOptions.interactivity.modes.bubble;e&&(t.retina.bubbleModeDistance=e.distance*t.retina.pixelRatio,void 0!==e.size&&(t.retina.bubbleModeSize=e.size*t.retina.pixelRatio))}async interact(t){const e=this.container.actualOptions.interactivity.events,i=e.onHover,s=e.onClick,o=i.enable,n=i.mode,a=s.enable,r=s.mode,c=e.onDiv;o&&Z(Vs,n)?this._hoverBubble():a&&Z(Vs,r)?this._clickBubble():nt(Vs,c,((e,i)=>this._singleSelectorHover(t,e,i)))}isEnabled(t){const e=this.container,i=e.actualOptions,s=e.interactivity.mouse,o=(t?.interactivity??i.interactivity).events,{onClick:n,onDiv:a,onHover:r}=o,c=ot(Vs,a);return!!(c||r.enable&&s.position||n.enable&&s.clickPosition)&&(Z(Vs,r.mode)||Z(Vs,n.mode)||c)}loadModeOptions(t,...e){t.bubble||(t.bubble=new qs);for(const i of e)t.bubble.load(i?.bubble)}reset(t){t.bubble.inRange=!1}}class Ws{constructor(){this.opacity=.5}load(t){t&&void 0!==t.opacity&&(this.opacity=t.opacity)}}class $s{constructor(){this.distance=80,this.links=new Ws,this.radius=60}load(t){t&&(void 0!==t.distance&&(this.distance=t.distance),this.links.load(t.links),void 0!==t.radius&&(this.radius=t.radius))}}function js(t,e,i,s){const o=t.actualOptions.interactivity.modes.connect;if(o)return function(t,e,i,s){const o=Math.floor(i.getRadius()/e.getRadius()),n=e.getFillColor(),a=i.getFillColor();if(!n||!a)return;const r=e.getPosition(),c=i.getPosition(),l=Vt(n,a,e.getRadius(),i.getRadius()),h=t.createLinearGradient(r.x,r.y,c.x,c.y);return h.addColorStop(0,Ht(n,s)),h.addColorStop(o>1?1:o,qt(l,s)),h.addColorStop(1,Ht(a,s)),h}(e,i,s,o.links.opacity)}function Gs(t,e,i){t.canvas.draw((s=>{const o=js(t,s,e,i);if(!o)return;const n=e.getPosition(),a=i.getPosition();!function(t,e,i,s,o){Nt(t,s,o),t.lineWidth=e,t.strokeStyle=i,t.stroke()}(s,e.retina.linksWidth??0,o,n,a)}))}class Ns extends Si{constructor(t){super(t)}clear(){}init(){const t=this.container,e=t.actualOptions.interactivity.modes.connect;e&&(t.retina.connectModeDistance=e.distance*t.retina.pixelRatio,t.retina.connectModeRadius=e.radius*t.retina.pixelRatio)}async interact(){const t=this.container;if(t.actualOptions.interactivity.events.onHover.enable&&"pointermove"===t.interactivity.status){const e=t.interactivity.mouse.position;if(!t.retina.connectModeDistance||t.retina.connectModeDistance<0||!t.retina.connectModeRadius||t.retina.connectModeRadius<0||!e)return;const i=Math.abs(t.retina.connectModeRadius),s=t.particles.quadTree.queryCircle(e,i,(t=>this.isEnabled(t)));let o=0;for(const e of s){const i=e.getPosition();for(const n of s.slice(o+1)){const s=n.getPosition(),o=Math.abs(t.retina.connectModeDistance),a=Math.abs(i.x-s.x),r=Math.abs(i.y-s.y);a{const n=e.getPosition();!function(t,e,i,s,o,n){Nt(t,i,s),t.strokeStyle=qt(o,n),t.lineWidth=e,t.stroke()}(t,e.retina.linksWidth??0,n,o,i,s)}))}class Qs extends Si{constructor(t){super(t)}clear(){}init(){const t=this.container,e=t.actualOptions.interactivity.modes.grab;e&&(t.retina.grabModeDistance=e.distance*t.retina.pixelRatio)}async interact(){const t=this.container,e=t.actualOptions.interactivity;if(!e.modes.grab||!e.events.onHover.enable||t.interactivity.status!==r)return;const i=t.interactivity.mouse.position;if(!i)return;const s=t.retina.grabModeDistance;if(!s||s<0)return;const o=t.particles.quadTree.queryCircle(i,s,(t=>this.isEnabled(t)));for(const n of o){const o=T(n.getPosition(),i);if(o>s)continue;const a=e.modes.grab.links,r=a.opacity,c=r-o*r/s;if(c<=0)continue;const l=a.color??n.options.links?.color;if(!t.particles.grabLineColor&&l){const i=e.modes.grab.links;t.particles.grabLineColor=Wt(l,i.blink,i.consent)}const h=Ut(n,void 0,t.particles.grabLineColor);h&&Zs(t,n,h,c,i)}}isEnabled(t){const e=this.container,i=e.interactivity.mouse,s=(t?.interactivity??e.actualOptions.interactivity).events;return s.onHover.enable&&!!i.position&&Z("grab",s.onHover.mode)}loadModeOptions(t,...e){t.grab||(t.grab=new Ys);for(const i of e)t.grab.load(i?.grab)}reset(){}}class Js extends Si{constructor(t){super(t),this.handleClickMode=t=>{if("pause"!==t)return;const e=this.container;e.getAnimationStatus()?e.pause():e.play()}}clear(){}init(){}async interact(){}isEnabled(){return!0}reset(){}}class Ks{constructor(){this.default=!0,this.groups=[],this.quantity=4}load(t){if(!t)return;void 0!==t.default&&(this.default=t.default),void 0!==t.groups&&(this.groups=t.groups.map((t=>t))),this.groups.length||(this.default=!0);const e=t.quantity;void 0!==e&&(this.quantity=S(e))}}class to extends Si{constructor(t){super(t),this.handleClickMode=t=>{if("push"!==t)return;const e=this.container,i=e.actualOptions.interactivity.modes.push;if(!i)return;const s=C(i.quantity);if(s<=0)return;const o=K([void 0,...i.groups]),n=void 0!==o?e.actualOptions.particles.groups[o]:void 0;e.particles.push(s,e.interactivity.mouse,n,o)}}clear(){}init(){}async interact(){}isEnabled(){return!0}loadModeOptions(t,...e){t.push||(t.push=new Ks);for(const i of e)t.push.load(i?.push)}reset(){}}class eo{constructor(){this.quantity=2}load(t){if(!t)return;const e=t.quantity;void 0!==e&&(this.quantity=S(e))}}class io extends Si{constructor(t){super(t),this.handleClickMode=t=>{const e=this.container,i=e.actualOptions;if(!i.interactivity.modes.remove||"remove"!==t)return;const s=C(i.interactivity.modes.remove.quantity);e.particles.removeQuantity(s)}}clear(){}init(){}async interact(){}isEnabled(){return!0}loadModeOptions(t,...e){t.remove||(t.remove=new eo);for(const i of e)t.remove.load(i?.remove)}reset(){}}class so{constructor(){this.distance=200,this.duration=.4,this.factor=100,this.speed=1,this.maxSpeed=50,this.easing="ease-out-quad"}load(t){t&&(void 0!==t.distance&&(this.distance=t.distance),void 0!==t.duration&&(this.duration=t.duration),void 0!==t.easing&&(this.easing=t.easing),void 0!==t.factor&&(this.factor=t.factor),void 0!==t.speed&&(this.speed=t.speed),void 0!==t.maxSpeed&&(this.maxSpeed=t.maxSpeed))}}class oo extends so{constructor(){super(),this.selectors=[]}load(t){super.load(t),t&&void 0!==t.selectors&&(this.selectors=t.selectors)}}class no extends so{load(t){super.load(t),t&&(this.divs=dt(t.divs,(t=>{const e=new oo;return e.load(t),e})))}}const ao="repulse";class ro extends Si{constructor(t,e){super(e),this._clickRepulse=()=>{const t=this.container,e=t.actualOptions.interactivity.modes.repulse;if(!e)return;const i=t.repulse||{particles:[]};if(i.finish||(i.count||(i.count=0),i.count++,i.count===t.particles.count&&(i.finish=!0)),i.clicking){const s=t.retina.repulseModeDistance;if(!s||s<0)return;const o=Math.pow(s/6,3),n=t.interactivity.mouse.clickPosition;if(void 0===n)return;const a=new yi(n.x,n.y,o),r=t.particles.quadTree.query(a,(t=>this.isEnabled(t)));for(const t of r){const{dx:s,dy:a,distance:r}=D(n,t.position),c=r**2,l=-o*e.speed/c;if(c<=o){i.particles.push(t);const e=y.create(s,a);e.length=l,t.velocity.setTo(e)}}}else if(!1===i.clicking){for(const t of i.particles)t.velocity.setTo(t.initialVelocity);i.particles=[]}},this._hoverRepulse=()=>{const t=this.container,e=t.interactivity.mouse.position,i=t.retina.repulseModeDistance;!i||i<0||!e||this._processRepulse(e,i,new yi(e.x,e.y,i))},this._processRepulse=(t,e,i,s)=>{const o=this.container,n=o.particles.quadTree.query(i,(t=>this.isEnabled(t))),a=o.actualOptions.interactivity.modes.repulse;if(!a)return;const{easing:r,speed:c,factor:l,maxSpeed:h}=a,d=w(r),u=(s?.speed??c)*l;for(const i of n){const{dx:s,dy:o,distance:n}=D(i.position,t),a=k(d(1-n/e)*u,0,h),r=y.create(0===n?u:s/n*a,0===n?u:o/n*a);i.position.addTo(r)}},this._singleSelectorRepulse=(t,e)=>{const i=this.container,s=i.actualOptions.interactivity.modes.repulse;if(!s)return;const o=document.querySelectorAll(t);o.length&&o.forEach((t=>{const o=t,n=i.retina.pixelRatio,a={x:(o.offsetLeft+o.offsetWidth/2)*n,y:(o.offsetTop+o.offsetHeight/2)*n},r=o.offsetWidth/2*n,c="circle"===e.type?new yi(a.x,a.y,r):new vi(o.offsetLeft*n,o.offsetTop*n,o.offsetWidth*n,o.offsetHeight*n),l=rt(s.divs,o);this._processRepulse(a,r,c,l)}))},this._engine=t,e.repulse||(e.repulse={particles:[]}),this.handleClickMode=t=>{const i=this.container.actualOptions.interactivity.modes.repulse;if(!i||t!==ao)return;e.repulse||(e.repulse={particles:[]});const s=e.repulse;s.clicking=!0,s.count=0;for(const t of e.repulse.particles)this.isEnabled(t)&&t.velocity.setTo(t.initialVelocity);s.particles=[],s.finish=!1,setTimeout((()=>{e.destroyed||(s.clicking=!1)}),1e3*i.duration)}}clear(){}init(){const t=this.container,e=t.actualOptions.interactivity.modes.repulse;e&&(t.retina.repulseModeDistance=e.distance*t.retina.pixelRatio)}async interact(){const t=this.container,e=t.actualOptions,i=t.interactivity.status===r,s=e.interactivity.events,o=s.onHover,n=o.enable,a=o.mode,c=s.onClick,l=c.enable,h=c.mode,d=s.onDiv;i&&n&&Z(ao,a)?this._hoverRepulse():l&&Z(ao,h)?this._clickRepulse():nt(ao,d,((t,e)=>this._singleSelectorRepulse(t,e)))}isEnabled(t){const e=this.container,i=e.actualOptions,s=e.interactivity.mouse,o=(t?.interactivity??i.interactivity).events,n=o.onDiv,a=o.onHover,r=o.onClick,c=ot(ao,n);if(!(c||a.enable&&s.position||r.enable&&s.clickPosition))return!1;const l=a.mode,h=r.mode;return Z(ao,l)||Z(ao,h)||c}loadModeOptions(t,...e){t.repulse||(t.repulse=new no);for(const i of e)t.repulse.load(i?.repulse)}reset(){}}class co{constructor(){this.factor=3,this.radius=200}load(t){t&&(void 0!==t.factor&&(this.factor=t.factor),void 0!==t.radius&&(this.radius=t.radius))}}class lo extends Si{constructor(t){super(t)}clear(t,e,i){t.slow.inRange&&!i||(t.slow.factor=1)}init(){const t=this.container,e=t.actualOptions.interactivity.modes.slow;e&&(t.retina.slowModeRadius=e.radius*t.retina.pixelRatio)}async interact(){}isEnabled(t){const e=this.container,i=e.interactivity.mouse,s=(t?.interactivity??e.actualOptions.interactivity).events;return s.onHover.enable&&!!i.position&&Z("slow",s.onHover.mode)}loadModeOptions(t,...e){t.slow||(t.slow=new co);for(const i of e)t.slow.load(i?.slow)}reset(t){t.slow.inRange=!1;const e=this.container,i=e.actualOptions,s=e.interactivity.mouse.position,o=e.retina.slowModeRadius,n=i.interactivity.modes.slow;if(!n||!o||o<0||!s)return;const a=T(s,t.getPosition()),r=a/o,c=n.factor,{slow:l}=t;a>o||(l.inRange=!0,l.factor=r/c)}}const ho=[0,4,2,1],uo=[8,8,4,2];class po{constructor(t){this.pos=0,this.data=new Uint8ClampedArray(t)}getString(t){const e=this.data.slice(this.pos,this.pos+t);return this.pos+=e.length,e.reduce(((t,e)=>t+String.fromCharCode(e)),"")}nextByte(){return this.data[this.pos++]}nextTwoBytes(){return this.pos+=2,this.data[this.pos-2]+(this.data[this.pos-1]<<8)}readSubBlocks(){let t="",e=0;do{e=this.data[this.pos++];for(let i=e;--i>=0;t+=String.fromCharCode(this.data[this.pos++]));}while(0!==e);return t}readSubBlocksBin(){let t=0,e=0;for(let i=0;0!==(t=this.data[this.pos+i]);i+=t+1)e+=t;const i=new Uint8Array(e);for(let e=0;0!==(t=this.data[this.pos++]);)for(let s=t;--s>=0;i[e++]=this.data[this.pos++]);return i}skipSubBlocks(){for(;0!==this.data[this.pos];this.pos+=this.data[this.pos]+1);this.pos++}}function fo(t,e){const i=[];for(let s=0;s>>3;const h=1<<1+(7&r);c&&(a.localColorTable=fo(t,h));const d=t=>{const{r:s,g:n,b:r}=(c?a.localColorTable:e.globalColorTable)[t];return{r:s,g:n,b:r,a:t===o(null)?i?~~((s+n+r)/3):0:255}},u=(()=>{try{return new ImageData(a.width,a.height,{colorSpace:"srgb"})}catch(t){if(t instanceof DOMException&&"IndexSizeError"===t.name)return null;throw t}})();if(null==u)throw new EvalError("GIF frame size is to large");const p=t.nextByte(),f=t.readSubBlocksBin(),v=1<{const i=t>>>3,s=7&t;return(f[i]+(f[i+1]<<8)+(f[i+2]<<16)&(1<>>s};if(l){for(let i=0,o=p+1,r=0,c=[[0]],l=0;l<4;l++){if(ho[l]=c.length?c.push(c[s].concat(c[s][0])):s!==v&&c.push(c[s].concat(c[i][0]));for(let s=0;s=a.height))break}n?.(t.pos/(t.data.length-1),s(!1)+1,u,{x:a.left,y:a.top},{width:e.width,height:e.height})}a.image=u,a.bitmap=await createImageBitmap(u)}else{for(let t=0,e=p+1,i=0,s=[[0]],o=-4;;){const n=t;if(t=y(i,e),i+=e,t===v){e=p+1,s.length=v+2;for(let t=0;t=s.length?s.push(s[n].concat(s[n][0])):n!==v&&s.push(s[n].concat(s[t][0]));for(let e=0;e=1<>>5,o.disposalMethod=(28&n)>>>2,o.userInputDelayFlag=2==(2&n);const a=1==(1&n);o.delayTime=10*t.nextTwoBytes();const r=t.nextByte();a&&s(r),t.pos++;break}case 255:{t.pos++;const i={identifier:t.getString(8),authenticationCode:t.getString(3),data:t.readSubBlocksBin()};e.applicationExtensions.push(i);break}case 254:e.comments.push([i(!1),t.readSubBlocks()]);break;case 1:if(0===e.globalColorTable.length)throw new EvalError("plain text extension without global color table");t.pos++,e.frames[i(!1)].plainTextData={left:t.nextTwoBytes(),top:t.nextTwoBytes(),width:t.nextTwoBytes(),height:t.nextTwoBytes(),charSize:{width:t.nextTwoBytes(),height:t.nextTwoBytes()},foregroundColor:t.nextByte(),backgroundColor:t.nextByte(),text:t.readSubBlocks()};break;default:t.skipSubBlocks()}}(t,e,s,o);break;default:throw new EvalError("undefined block found")}return!1}const yo=/(#(?:[0-9a-f]{2}){2,4}|(#[0-9a-f]{3})|(rgb|hsl)a?\((-?\d+%?[,\s]+){2,3}\s*[\d.]+%?\))|currentcolor/gi;async function mo(t){return new Promise((e=>{t.loading=!0;const i=new Image;t.element=i,i.addEventListener("load",(()=>{t.loading=!1,e()})),i.addEventListener("error",(()=>{t.element=void 0,t.error=!0,t.loading=!1,W().error(`${f} loading image: ${t.source}`),e()})),i.src=t.source}))}async function go(t){if("gif"===t.type){t.loading=!0;try{t.gifData=await async function(t,e,i){i||(i=!1);const s=await fetch(t);if(!s.ok&&404===s.status)throw new EvalError("file not found");const o=await s.arrayBuffer(),n={width:0,height:0,totalTime:0,colorRes:0,pixelAspectRatio:0,frames:[],sortFlag:!1,globalColorTable:[],backgroundImage:new ImageData(1,1,{colorSpace:"srgb"}),comments:[],applicationExtensions:[]},a=new po(new Uint8ClampedArray(o));if("GIF89a"!==a.getString(6))throw new Error("not a supported GIF file");n.width=a.nextTwoBytes(),n.height=a.nextTwoBytes();const r=a.nextByte(),c=128==(128&r);n.colorRes=(112&r)>>>4,n.sortFlag=8==(8&r);const l=1<<1+(7&r),h=a.nextByte();n.pixelAspectRatio=a.nextByte(),0!==n.pixelAspectRatio&&(n.pixelAspectRatio=(n.pixelAspectRatio+15)/64),c&&(n.globalColorTable=fo(a,l));const d=(()=>{try{return new ImageData(n.width,n.height,{colorSpace:"srgb"})}catch(t){if(t instanceof DOMException&&"IndexSizeError"===t.name)return null;throw t}})();if(null==d)throw new Error("GIF frame size is to large");const{r:u,g:p,b:f}=n.globalColorTable[h];d.data.set(c?[u,p,f,255]:[0,0,0,0]);for(let t=4;t(t&&(y=!0),v),b=t=>(null!=t&&(m=t),m);try{do{y&&(n.frames.push({left:0,top:0,width:0,height:0,disposalMethod:0,image:new ImageData(1,1,{colorSpace:"srgb"}),plainTextData:null,userInputDelayFlag:!1,delayTime:0,sortFlag:!1,localColorTable:[],reserved:0,GCreserved:0}),v++,m=-1,y=!1)}while(!await vo(a,n,i,g,b,e));n.frames.length--;for(const t of n.frames){if(t.userInputDelayFlag&&0===t.delayTime){n.totalTime=1/0;break}n.totalTime+=t.delayTime}return n}catch(t){if(t instanceof EvalError)throw new Error(`error while parsing frame ${v} "${t.message}"`);throw t}}(t.source),t.gifLoopCount=function(t){for(const e of t.applicationExtensions)if(e.identifier+e.authenticationCode==="NETSCAPE2.0")return e.data[1]+(e.data[2]<<8);return NaN}(t.gifData)??0,0===t.gifLoopCount&&(t.gifLoopCount=1/0)}catch{t.error=!0}t.loading=!1}else await mo(t)}async function bo(t){if("svg"!==t.type)return void await mo(t);t.loading=!0;const e=await fetch(t.source);e.ok?t.svgData=await e.text():(W().error(`${f} Image not found`),t.error=!0),t.loading=!1}function wo(t,e,i,s){const o=function(t,e,i){const{svgData:s}=t;if(!s)return"";const o=Ht(e,i);if(s.includes("fill"))return s.replace(yo,(()=>o));const n=s.indexOf(">");return`${s.substring(0,n)} fill="${o}"${s.substring(n)}`}(t,i,s.opacity?.value??1),n={color:i,gif:e.gif,data:{...t,svgData:o},loaded:!1,ratio:e.width/e.height,replaceColor:e.replaceColor,source:e.src};return new Promise((e=>{const i=new Blob([o],{type:"image/svg+xml"}),s=URL||window.URL||window.webkitURL||window,a=s.createObjectURL(i),r=new Image;r.addEventListener("load",(()=>{n.loaded=!0,n.element=r,e(n),s.revokeObjectURL(a)})),r.addEventListener("error",(async()=>{s.revokeObjectURL(a);const i={...t,error:!1,loading:!0};await mo(i),n.loaded=!0,n.element=i.element,e(n)})),r.src=a}))}class xo{constructor(t){this.loadImageShape=async t=>{if(!this._engine.loadImage)throw new Error(`${f} image shape not initialized`);await this._engine.loadImage({gif:t.gif,name:t.name,replaceColor:t.replaceColor??!1,src:t.src})},this._engine=t}addImage(t){this._engine.images||(this._engine.images=[]),this._engine.images.push(t)}draw(t){const{context:e,radius:i,particle:s,opacity:o,delta:n}=t,a=s.image,r=a?.element;if(a){if(e.globalAlpha=o,a.gif&&a.gifData){const t=new OffscreenCanvas(a.gifData.width,a.gifData.height),o=t.getContext("2d");if(!o)throw new Error("could not create offscreen canvas context");o.imageSmoothingQuality="low",o.imageSmoothingEnabled=!1,o.clearRect(0,0,t.width,t.height),void 0===s.gifLoopCount&&(s.gifLoopCount=a.gifLoopCount??0);let r=s.gifFrame??0;const c={x:.5*-a.gifData.width,y:.5*-a.gifData.height},l=a.gifData.frames[r];if(void 0===s.gifTime&&(s.gifTime=0),!l.bitmap)return;switch(e.scale(i/a.gifData.width,i/a.gifData.height),l.disposalMethod){case 4:case 5:case 6:case 7:case 0:o.drawImage(l.bitmap,l.left,l.top),e.drawImage(t,c.x,c.y),o.clearRect(0,0,t.width,t.height);break;case 1:o.drawImage(l.bitmap,l.left,l.top),e.drawImage(t,c.x,c.y);break;case 2:o.drawImage(l.bitmap,l.left,l.top),e.drawImage(t,c.x,c.y),o.clearRect(0,0,t.width,t.height),0===a.gifData.globalColorTable.length?o.putImageData(a.gifData.frames[0].image,c.x+l.left,c.y+l.top):o.putImageData(a.gifData.backgroundImage,c.x,c.y);break;case 3:{const i=o.getImageData(0,0,t.width,t.height);o.drawImage(l.bitmap,l.left,l.top),e.drawImage(t,c.x,c.y),o.clearRect(0,0,t.width,t.height),o.putImageData(i,0,0)}}if(s.gifTime+=n.value,s.gifTime>l.delayTime){if(s.gifTime-=l.delayTime,++r>=a.gifData.frames.length){if(--s.gifLoopCount<=0)return;r=0,o.clearRect(0,0,t.width,t.height)}s.gifFrame=r}e.scale(a.gifData.width/i,a.gifData.height/i)}else if(r){const t=a.ratio,s={x:-i,y:-i},o=2*i;e.drawImage(r,s.x,s.y,o,o/t)}e.globalAlpha=1}}getSidesCount(){return 12}async init(t){const e=t.actualOptions;if(e.preload&&this._engine.loadImage)for(const t of e.preload)await this._engine.loadImage(t)}loadShape(t){if("image"!==t.shape&&"images"!==t.shape)return;this._engine.images||(this._engine.images=[]);const e=t.shapeData;if(!e)return;this._engine.images.find((t=>t.name===e.name||t.source===e.src))||this.loadImageShape(e).then((()=>{this.loadShape(t)}))}particleInit(t,e){if("image"!==e.shape&&"images"!==e.shape)return;this._engine.images||(this._engine.images=[]);const i=this._engine.images,s=e.shapeData;if(!s)return;const o=e.getFillColor(),n=i.find((t=>t.name===s.name||t.source===s.src));if(!n)return;const a=s.replaceColor??n.replaceColor;n.loading?setTimeout((()=>{this.particleInit(t,e)})):(async()=>{let t;t=n.svgData&&o?await wo(n,s,o,e):{color:o,data:n,element:n.element,gif:n.gif,gifData:n.gifData,gifLoopCount:n.gifLoopCount,loaded:!0,ratio:s.width&&s.height?s.width/s.height:n.ratio??1,replaceColor:a,source:s.src},t.ratio||(t.ratio=1);const i={image:t,fill:s.fill??e.shapeFill,close:s.close??e.shapeClose};e.image=i.image,e.shapeFill=i.fill,e.shapeClose=i.close})()}}class _o{constructor(){this.src="",this.gif=!1}load(t){t&&(void 0!==t.gif&&(this.gif=t.gif),void 0!==t.height&&(this.height=t.height),void 0!==t.name&&(this.name=t.name),void 0!==t.replaceColor&&(this.replaceColor=t.replaceColor),void 0!==t.src&&(this.src=t.src),void 0!==t.width&&(this.width=t.width))}}class ko{constructor(t){this.id="imagePreloader",this._engine=t}getPlugin(){return{}}loadOptions(t,e){if(!e||!e.preload)return;t.preload||(t.preload=[]);const i=t.preload;for(const t of e.preload){const e=i.find((e=>e.name===t.name||e.src===t.src));if(e)e.load(t);else{const e=new _o;e.load(t),i.push(e)}}}needsPlugin(){return!0}}async function zo(t,e=!0){!function(t){t.loadImage||(t.loadImage=async e=>{if(!e.name&&!e.src)throw new Error(`${f} no image source provided`);if(t.images||(t.images=[]),!t.images.find((t=>t.name===e.name||t.source===e.src)))try{const i={gif:e.gif??!1,name:e.name??e.src,source:e.src,type:e.src.substring(e.src.length-3),error:!1,loading:!0,replaceColor:e.replaceColor,ratio:e.width&&e.height?e.width/e.height:void 0};t.images.push(i);const s=e.gif?go:e.replaceColor?bo:mo;await s(i)}catch{throw new Error(`${f} ${e.name??e.src} not found`)}})}(t);const i=new ko(t);await t.addPlugin(i,e),await t.addShape(["image","images"],new xo(t),e)}class Mo extends Re{constructor(){super(),this.sync=!1}load(t){t&&(super.load(t),void 0!==t.sync&&(this.sync=t.sync))}}class Co extends Re{constructor(){super(),this.sync=!1}load(t){t&&(super.load(t),void 0!==t.sync&&(this.sync=t.sync))}}class Po{constructor(){this.count=0,this.delay=new Mo,this.duration=new Co}load(t){t&&(void 0!==t.count&&(this.count=t.count),this.delay.load(t.delay),this.duration.load(t.duration))}}class Oo{constructor(t){this.container=t}init(t){const e=this.container,i=t.options.life;i&&(t.life={delay:e.retina.reduceFactor?C(i.delay.value)*(i.delay.sync?1:_())/e.retina.reduceFactor*1e3:0,delayTime:0,duration:e.retina.reduceFactor?C(i.duration.value)*(i.duration.sync?1:_())/e.retina.reduceFactor*1e3:0,time:0,count:i.count},t.life.duration<=0&&(t.life.duration=-1),t.life.count<=0&&(t.life.count=-1),t.life&&(t.spawning=t.life.delay>0))}isEnabled(t){return!t.destroyed}loadOptions(t,...e){t.life||(t.life=new Po);for(const i of e)t.life.load(i?.life)}update(t,e){if(!this.isEnabled(t)||!t.life)return;const i=t.life;let s=!1;if(t.spawning){if(i.delayTime+=e.value,!(i.delayTime>=t.life.delay))return;s=!0,t.spawning=!1,i.delayTime=0,i.time=0}if(-1===i.duration)return;if(t.spawning)return;if(s?i.time=0:i.time+=e.value,i.time0&&t.life.count--,0===t.life.count)return void t.destroy();const o=this.container.canvas.size,n=S(0,o.width),a=S(0,o.width);t.position.x=M(n),t.position.y=M(a),t.spawning=!0,i.delayTime=0,i.time=0,t.reset();const r=t.options.life;r&&(i.delay=1e3*C(r.delay.value),i.duration=1e3*C(r.duration.value))}}class So{draw(t){const{context:e,particle:i,radius:s}=t,o=i.shapeData;e.moveTo(-s/2,0),e.lineTo(s/2,0),e.lineCap=o?.cap??"butt"}getSidesCount(){return 1}}class Do{init(){}isEnabled(t){return!j()&&!t.destroyed&&t.container.actualOptions.interactivity.events.onHover.parallax.enable}move(t){const e=t.container,i=e.actualOptions.interactivity.events.onHover.parallax;if(j()||!i.enable)return;const s=i.force,o=e.interactivity.mouse.position;if(!o)return;const n=e.canvas.size,a=.5*n.width,r=.5*n.height,c=i.smooth,l=t.getRadius()/s,h=(o.x-a)*l,d=(o.y-r)*l,{offset:u}=t;u.x+=(h-u.x)/c,u.y+=(d-u.y)/c}}class To extends Di{constructor(t){super(t)}clear(){}init(){}async interact(t){const e=this.container;void 0===t.attractDistance&&(t.attractDistance=C(t.options.move.attract.distance)*e.retina.pixelRatio);const i=t.attractDistance,s=t.getPosition(),o=e.particles.quadTree.queryCircle(s,i);for(const e of o){if(t===e||!e.options.move.attract.enable||e.destroyed||e.spawning)continue;const i=e.getPosition(),{dx:o,dy:n}=D(s,i),a=t.options.move.attract.rotate,r=o/(1e3*a.x),c=n/(1e3*a.y),l=e.size.value/t.size.value,h=1/l;t.velocity.x-=r*l,t.velocity.y-=c*l,e.velocity.x+=r*h,e.velocity.y+=c*h}}isEnabled(t){return t.options.move.attract.enable}reset(){}}function Ro(t,e,i,s,o,n){const a=k(t.options.collisions.absorb.speed*o.factor/10,0,s);t.size.value+=a/2,i.size.value-=a,s<=n&&(i.size.value=0,i.destroy())}const Eo=t=>{void 0===t.collisionMaxSpeed&&(t.collisionMaxSpeed=C(t.options.collisions.maxSpeed)),t.velocity.length>t.collisionMaxSpeed&&(t.velocity.length=t.collisionMaxSpeed)};function Io(t,e){lt(ct(t),ct(e)),Eo(t),Eo(e)}function Lo(t,e,i,s){switch(t.options.collisions.mode){case"absorb":!function(t,e,i,s){const o=t.getRadius(),n=e.getRadius();void 0===o&&void 0!==n?t.destroy():void 0!==o&&void 0===n?e.destroy():void 0!==o&&void 0!==n&&(o>=n?Ro(t,0,e,n,i,s):Ro(e,0,t,o,i,s))}(t,e,i,s);break;case"bounce":Io(t,e);break;case"destroy":!function(t,e){t.unbreakable||e.unbreakable||Io(t,e),void 0===t.getRadius()&&void 0!==e.getRadius()?t.destroy():void 0!==t.getRadius()&&void 0===e.getRadius()?e.destroy():void 0!==t.getRadius()&&void 0!==e.getRadius()&&(t.getRadius()>=e.getRadius()?e:t).destroy()}(t,e)}}class Ao extends Di{constructor(t){super(t)}clear(){}init(){}async interact(t,e){if(t.destroyed||t.spawning)return;const i=this.container,s=t.getPosition(),o=t.getRadius(),n=i.particles.quadTree.queryCircle(s,2*o);for(const a of n){if(t===a||!a.options.collisions.enable||t.options.collisions.mode!==a.options.collisions.mode||a.destroyed||a.spawning)continue;const n=a.getPosition(),r=a.getRadius();if(Math.abs(Math.round(s.z)-Math.round(n.z))>o+r)continue;T(s,n)>o+r||Lo(t,a,e,i.retina.pixelRatio)}}isEnabled(t){return t.options.collisions.enable}reset(){}}class Fo extends yi{constructor(t,e,i,s){super(t,e,i),this.canvasSize=s,this.canvasSize={...s}}contains(t){const{width:e,height:i}=this.canvasSize,{x:s,y:o}=t;return super.contains(t)||super.contains({x:s-e,y:o})||super.contains({x:s-e,y:o-i})||super.contains({x:s,y:o-i})}intersects(t){if(super.intersects(t))return!0;const e=t,i=t,s={x:t.position.x-this.canvasSize.width,y:t.position.y-this.canvasSize.height};if(void 0!==i.radius){const t=new yi(s.x,s.y,2*i.radius);return super.intersects(t)}if(void 0!==e.size){const t=new vi(s.x,s.y,2*e.size.width,2*e.size.height);return super.intersects(t)}return!1}}class Bo{constructor(){this.blur=5,this.color=new ce,this.color.value="#000",this.enable=!1}load(t){t&&(void 0!==t.blur&&(this.blur=t.blur),this.color=ce.create(this.color,t.color),void 0!==t.enable&&(this.enable=t.enable))}}class qo{constructor(){this.enable=!1,this.frequency=1}load(t){t&&(void 0!==t.color&&(this.color=ce.create(this.color,t.color)),void 0!==t.enable&&(this.enable=t.enable),void 0!==t.frequency&&(this.frequency=t.frequency),void 0!==t.opacity&&(this.opacity=t.opacity))}}class Ho{constructor(){this.blink=!1,this.color=new ce,this.color.value="#fff",this.consent=!1,this.distance=100,this.enable=!1,this.frequency=1,this.opacity=1,this.shadow=new Bo,this.triangles=new qo,this.width=1,this.warp=!1}load(t){t&&(void 0!==t.id&&(this.id=t.id),void 0!==t.blink&&(this.blink=t.blink),this.color=ce.create(this.color,t.color),void 0!==t.consent&&(this.consent=t.consent),void 0!==t.distance&&(this.distance=t.distance),void 0!==t.enable&&(this.enable=t.enable),void 0!==t.frequency&&(this.frequency=t.frequency),void 0!==t.opacity&&(this.opacity=t.opacity),this.shadow.load(t.shadow),this.triangles.load(t.triangles),void 0!==t.width&&(this.width=t.width),void 0!==t.warp&&(this.warp=t.warp))}}function Vo(t,e,i,s,o){const{dx:n,dy:a,distance:r}=D(t,e);if(!o||r<=i)return r;const c={x:Math.abs(n),y:Math.abs(a)},l=Math.min(c.x,s.width-c.x),h=Math.min(c.y,s.height-c.y);return Math.sqrt(l**2+h**2)}class Uo extends Di{constructor(t){super(t),this._setColor=t=>{if(!t.options.links)return;const e=this.linkContainer,i=t.options.links;let s=void 0===i.id?e.particles.linksColor:e.particles.linksColors.get(i.id);if(s)return;s=Wt(i.color,i.blink,i.consent),void 0===i.id?e.particles.linksColor=s:e.particles.linksColors.set(i.id,s)},this.linkContainer=t}clear(){}init(){this.linkContainer.particles.linksColor=void 0,this.linkContainer.particles.linksColors=new Map}async interact(t){if(!t.options.links)return;t.links=[];const e=t.getPosition(),i=this.container,s=i.canvas.size;if(e.x<0||e.y<0||e.x>s.width||e.y>s.height)return;const o=t.options.links,n=o.opacity,a=t.retina.linksDistance??0,r=o.warp,c=r?new Fo(e.x,e.y,a,s):new yi(e.x,e.y,a),l=i.particles.quadTree.query(c);for(const i of l){const c=i.options.links;if(t===i||!c?.enable||o.id!==c.id||i.spawning||i.destroyed||!i.links||t.links.some((t=>t.destination===i))||i.links.some((e=>e.destination===t)))continue;const l=i.getPosition();if(l.x<0||l.y<0||l.x>s.width||l.y>s.height)continue;const h=Vo(e,l,a,s,r&&c.warp);if(h>a)continue;const d=(1-h/a)*n;this._setColor(t),t.links.push({destination:i,opacity:d})}}isEnabled(t){return!!t.options.links?.enable}loadParticlesOptions(t,...e){t.links||(t.links=new Ho);for(const i of e)t.links.load(i?.links)}reset(){}}function Wo(t,e){const i=((s=t.map((t=>t.id))).sort(((t,e)=>t-e)),s.join("_"));var s;let o=e.get(i);return void 0===o&&(o=_(),e.set(i,o)),o}class $o{constructor(t){this.container=t,this._drawLinkLine=(t,e)=>{const i=t.options.links;if(!i?.enable)return;const s=this.container,o=s.actualOptions,n=e.destination,a=t.getPosition(),r=n.getPosition();let c=e.opacity;s.canvas.draw((e=>{let l;const h=t.options.twinkle?.lines;if(h?.enable){const t=h.frequency,e=St(h.color);_(){const s=t.options.links;if(!s?.enable)return;const o=s.triangles;if(!o.enable)return;const n=this.container,a=n.actualOptions,r=e.destination,c=i.destination,l=o.opacity??(e.opacity+i.opacity)/2;l<=0||n.canvas.draw((e=>{const i=t.getPosition(),h=r.getPosition(),d=c.getPosition(),u=t.retina.linksDistance??0;if(T(i,h)>u||T(d,h)>u||T(d,i)>u)return;let p=St(o.color);if(!p){const e=void 0!==s.id?n.particles.linksColors.get(s.id):n.particles.linksColor;p=Ut(t,r,e)}p&&function(t){const{context:e,pos1:i,pos2:s,pos3:o,backgroundMask:n,colorTriangle:a,opacityTriangle:r}=t;!function(t,e,i,s){t.beginPath(),t.moveTo(e.x,e.y),t.lineTo(i.x,i.y),t.lineTo(s.x,s.y),t.closePath()}(e,i,s,o),n.enable&&(e.globalCompositeOperation=n.composite),e.fillStyle=qt(a,r),e.fill()}({context:e,pos1:i,pos2:h,pos3:d,backgroundMask:a.backgroundMask,colorTriangle:p,opacityTriangle:l})}))},this._drawTriangles=(t,e,i,s)=>{const o=i.destination;if(!t.links?.triangles.enable||!o.options.links?.triangles.enable)return;const n=o.links?.filter((t=>{const e=this._getLinkFrequency(o,t.destination);return o.options.links&&e<=o.options.links.frequency&&s.findIndex((e=>e.destination===t.destination))>=0}));if(n?.length)for(const s of n){const n=s.destination;this._getTriangleFrequency(e,o,n)>t.links.triangles.frequency||this._drawLinkTriangle(e,i,s)}},this._getLinkFrequency=(t,e)=>Wo([t,e],this._freqs.links),this._getTriangleFrequency=(t,e,i)=>Wo([t,e,i],this._freqs.triangles),this._freqs={links:new Map,triangles:new Map}}drawParticle(t,e){const{links:i,options:s}=e;if(!i||i.length<=0)return;const o=i.filter((t=>s.links&&this._getLinkFrequency(e,t.destination)<=s.links.frequency));for(const t of o)this._drawTriangles(s,e,t,o),t.opacity>0&&(e.retina.linksWidth??0)>0&&this._drawLinkLine(e,t)}async init(){this._freqs.links=new Map,this._freqs.triangles=new Map}particleCreated(t){if(t.links=[],!t.options.links)return;const e=this.container.retina.pixelRatio,{retina:i}=t,{distance:s,width:o}=t.options.links;i.linksDistance=s*e,i.linksWidth=o*e}particleDestroyed(t){t.links=[]}}class jo{constructor(){this.id="links"}getPlugin(t){return new $o(t)}loadOptions(){}needsPlugin(){return!0}}async function Go(t,e=!0){await async function(t,e=!0){await t.addInteractor("particlesLinks",(t=>new Uo(t)),e)}(t,e),await async function(t,e=!0){const i=new jo;await t.addPlugin(i,e)}(t,e)}class No{draw(t){const{context:e,particle:i,radius:s}=t,o=this.getCenter(i,s),n=this.getSidesData(i,s),a=n.count.numerator*n.count.denominator,r=n.count.numerator/n.count.denominator,c=180*(r-2)/r,l=Math.PI-Math.PI*c/180;if(e){e.beginPath(),e.translate(o.x,o.y),e.moveTo(0,0);for(let t=0;t0?"counter-clockwise":"clockwise"}switch(i){case"counter-clockwise":case"counterClockwise":t.rotate.status="decreasing";break;case"clockwise":t.rotate.status="increasing"}const s=e.animation;s.enable&&(t.rotate.decay=1-C(s.decay),t.rotate.velocity=C(s.speed)/360*this.container.retina.reduceFactor,s.sync||(t.rotate.velocity*=_())),t.rotation=t.rotate.value}isEnabled(t){const e=t.options.rotate;return!!e&&(!t.destroyed&&!t.spawning&&e.animation.enable&&!e.path)}loadOptions(t,...e){t.rotate||(t.rotate=new Jo);for(const i of e)t.rotate.load(i?.rotate)}update(t,e){this.isEnabled(t)&&(!function(t,e){const i=t.rotate,s=t.options.rotate;if(!i||!s)return;const o=s.animation,n=(i.velocity??0)*e.factor,a=2*Math.PI,r=i.decay??1;o.enable&&("increasing"===i.status?(i.value+=n,i.value>a&&(i.value-=a)):(i.value-=n,i.value<0&&(i.value+=a)),i.velocity&&1!==r&&(i.velocity*=r))}(t,e),t.rotation=t.rotate?.value??0)}}const tn=Math.sqrt(2);class en{draw(t){const{context:e,radius:i}=t,s=i/tn,o=2*s;e.rect(-s,-s,o,o)}getSidesCount(){return 4}}class sn{draw(t){const{context:e,particle:i,radius:s}=t,o=i.sides,n=i.starInset??2;e.moveTo(0,0-s);for(let t=0;t0&&(e.loops??0)>(e.maxLoops??0))return;if(e.time||(e.time=0),(e.delayTime??0)>0&&e.time<(e.delayTime??0)&&(e.time+=t.value),(e.delayTime??0)>0&&e.time<(e.delayTime??0))return;const n=M(i.offset),a=(e.velocity??0)*t.factor+3.6*n,r=e.decay??1;o&&"increasing"!==e.status?(e.value-=a,e.value<0&&(e.loops||(e.loops=0),e.loops++,e.status="increasing",e.value+=e.value)):(e.value+=a,e.value>s&&(e.loops||(e.loops=0),e.loops++,o&&(e.status="decreasing",e.value-=e.value%s))),e.velocity&&1!==r&&(e.velocity*=r),e.value>s&&(e.value%=s)}class nn{constructor(t){this.container=t}init(t){const e=this.container,i=t.options,s=ut(i.stroke,t.id,i.reduceDuplicates);t.strokeWidth=C(s.width)*e.retina.pixelRatio,t.strokeOpacity=C(s.opacity??1),t.strokeAnimation=s.color?.animation;const o=Rt(s.color)??t.getFillColor();o&&(t.strokeColor=jt(o,t.strokeAnimation,e.retina.reduceFactor))}isEnabled(t){const e=t.strokeAnimation,{strokeColor:i}=t;return!t.destroyed&&!t.spawning&&!!e&&(void 0!==i?.h.value&&i.h.enable||void 0!==i?.s.value&&i.s.enable||void 0!==i?.l.value&&i.l.enable)}update(t,e){this.isEnabled(t)&&function(t,e){if(!t.strokeColor||!t.strokeAnimation)return;const{h:i,s,l:o}=t.strokeColor,{h:n,s:a,l:r}=t.strokeAnimation;i&&on(e,i,n,360,!1),s&&on(e,s,a,100,!0),o&&on(e,o,r,100,!0)}(t,e)}}async function an(t,e=!0){await async function(t,e=!0){await t.addMover("parallax",(()=>new Do),e)}(t,!1),await async function(t,e=!0){await t.addInteractor("externalAttract",(e=>new Es(t,e)),e)}(t,!1),await async function(t,e=!0){await t.addInteractor("externalBounce",(t=>new As(t)),e)}(t,!1),await async function(t,e=!0){await t.addInteractor("externalBubble",(t=>new Us(t)),e)}(t,!1),await async function(t,e=!0){await t.addInteractor("externalConnect",(t=>new Ns(t)),e)}(t,!1),await async function(t,e=!0){await t.addInteractor("externalGrab",(t=>new Qs(t)),e)}(t,!1),await async function(t,e=!0){await t.addInteractor("externalPause",(t=>new Js(t)),e)}(t,!1),await async function(t,e=!0){await t.addInteractor("externalPush",(t=>new to(t)),e)}(t,!1),await async function(t,e=!0){await t.addInteractor("externalRemove",(t=>new io(t)),e)}(t,!1),await async function(t,e=!0){await t.addInteractor("externalRepulse",(e=>new ro(t,e)),e)}(t,!1),await async function(t,e=!0){await t.addInteractor("externalSlow",(t=>new lo(t)),e)}(t,!1),await async function(t,e=!0){await t.addInteractor("particlesAttract",(t=>new To(t)),e)}(t,!1),await async function(t,e=!0){await t.addInteractor("particlesCollisions",(t=>new Ao(t)),e)}(t,!1),await Go(t,!1),await async function(){b("ease-in-quad",(t=>t**2)),b("ease-out-quad",(t=>1-(1-t)**2)),b("ease-in-out-quad",(t=>t<.5?2*t**2:1-(-2*t+2)**2/2))}(),await async function(t,e=!0){await t.addShape(Os,new Ds,e)}(t,!1),await zo(t,!1),await async function(t,e=!0){await t.addShape("line",new So,e)}(t,!1),await Zo(t,!1),await async function(t,e=!0){await t.addShape(["edge","square"],new en,e)}(t,!1),await async function(t,e=!0){await t.addShape("star",new sn,e)}(t,!1),await async function(t,e=!0){await t.addParticleUpdater("life",(t=>new Oo(t)),e)}(t,!1),await async function(t,e=!0){await t.addParticleUpdater("rotate",(t=>new Ko(t)),e)}(t,!1),await async function(t,e=!0){await t.addParticleUpdater("strokeColor",(t=>new nn(t)),e)}(t,!1),await Ps(t,e)}const rn=["text","character","char","multiline-text"];class cn{constructor(){this._drawLine=(t,e,i,s,o,n)=>{const a={x:-(e.length*i/2),y:i/2},r=2*i;n?t.fillText(e,a.x,a.y+r*o):t.strokeText(e,a.x,a.y+r*o)}}draw(t){const{context:e,particle:i,radius:s,opacity:o}=t,n=i.shapeData;if(!n)return;const a=n.value;if(void 0===a)return;void 0===i.text&&(i.text=ut(a,i.randomIndexData));const r=i.text,c=n.style??"",l=n.weight??"400",h=2*Math.round(s),d=n.font??"Verdana",u=i.shapeFill,p=r?.split("\n");if(p){e.font=`${c} ${l} ${h}px "${d}"`,e.globalAlpha=o;for(let t=0;tZ(t,e.particles.shape.type)))){const t=rn.map((t=>e.particles.shape.options[t])).find((t=>!!t)),i=[];dt(t,(t=>{i.push(Q(t.font,t.weight))})),await Promise.all(i)}}particleInit(t,e){if(!e.shape||!rn.includes(e.shape))return;const i=e.shapeData;if(void 0===i)return;const s=i.value;void 0!==s&&(e.text=ut(s,e.randomIndexData))}}class ln{constructor(){this.enable=!1,this.speed=0,this.decay=0,this.sync=!1}load(t){t&&(void 0!==t.enable&&(this.enable=t.enable),void 0!==t.speed&&(this.speed=S(t.speed)),void 0!==t.decay&&(this.decay=S(t.decay)),void 0!==t.sync&&(this.sync=t.sync))}}class hn extends Re{constructor(){super(),this.animation=new ln,this.direction="clockwise",this.enable=!1,this.value=0}load(t){super.load(t),t&&(this.animation.load(t.animation),void 0!==t.direction&&(this.direction=t.direction),void 0!==t.enable&&(this.enable=t.enable))}}class dn{constructor(t){this.container=t}getTransformValues(t){const e=t.tilt?.enable&&t.tilt;return{b:e?Math.cos(e.value)*e.cosDirection:void 0,c:e?Math.sin(e.value)*e.sinDirection:void 0}}init(t){const e=t.options.tilt;if(!e)return;t.tilt={enable:e.enable,value:C(e.value)*Math.PI/180,sinDirection:_()>=.5?1:-1,cosDirection:_()>=.5?1:-1};let i=e.direction;if("random"===i){i=Math.floor(2*_())>0?"counter-clockwise":"clockwise"}switch(i){case"counter-clockwise":case"counterClockwise":t.tilt.status="decreasing";break;case"clockwise":t.tilt.status="increasing"}const s=t.options.tilt?.animation;s?.enable&&(t.tilt.decay=1-C(s.decay),t.tilt.velocity=C(s.speed)/360*this.container.retina.reduceFactor,s.sync||(t.tilt.velocity*=_()))}isEnabled(t){const e=t.options.tilt?.animation;return!t.destroyed&&!t.spawning&&!!e?.enable}loadOptions(t,...e){t.tilt||(t.tilt=new hn);for(const i of e)t.tilt.load(i?.tilt)}update(t,e){this.isEnabled(t)&&function(t,e){if(!t.tilt||!t.options.tilt)return;const i=t.options.tilt.animation,s=(t.tilt.velocity??0)*e.factor,o=2*Math.PI,n=t.tilt.decay??1;i.enable&&("increasing"===t.tilt.status?(t.tilt.value+=s,t.tilt.value>o&&(t.tilt.value-=o)):(t.tilt.value-=s,t.tilt.value<0&&(t.tilt.value+=o)),t.tilt.velocity&&1!==n&&(t.tilt.velocity*=n))}(t,e)}}class un{constructor(){this.enable=!1,this.frequency=.05,this.opacity=1}load(t){t&&(void 0!==t.color&&(this.color=ce.create(this.color,t.color)),void 0!==t.enable&&(this.enable=t.enable),void 0!==t.frequency&&(this.frequency=t.frequency),void 0!==t.opacity&&(this.opacity=S(t.opacity)))}}class pn{constructor(){this.lines=new un,this.particles=new un}load(t){t&&(this.lines.load(t.lines),this.particles.load(t.particles))}}class fn{getColorStyles(t,e,i,s){const o=t.options.twinkle;if(!o)return{};const n=o.particles,a=n.enable&&_()a&&(s.angle-=a),r.x+=n*Math.cos(s.angle),r.y+=n*Math.abs(Math.sin(s.angle))}(t,e)}}async function gn(t,e=!0){await async function(t,e=!0){await t.addParticleUpdater("destroy",(e=>new $i(t,e)),e)}(t,!1),await async function(t,e=!0){await t.addParticleUpdater("roll",(()=>new fs),e)}(t,!1),await async function(t,e=!0){await t.addParticleUpdater("tilt",(t=>new dn(t)),e)}(t,!1),await async function(t,e=!0){await t.addParticleUpdater("twinkle",(()=>new fn),e)}(t,!1),await async function(t,e=!0){await t.addParticleUpdater("wobble",(t=>new mn(t)),e)}(t,!1),await async function(t,e=!0){await t.addShape(rn,new cn,e)}(t,!1),await async function(t,e=!0){await t.addInteractor("externalTrail",(t=>new ds(t)),e)}(t,!1),await async function(t,e=!0){await t.addPlugin(new Fi,e)}(t,!1),await async function(t,e=!0){t.emitterShapeManager||(t.emitterShapeManager=new es(t)),t.addEmitterShapeGenerator||(t.addEmitterShapeGenerator=(e,i)=>{t.emitterShapeManager?.addShapeGenerator(e,i)});const i=new ss(t);await t.addPlugin(i,e)}(t,!1),await async function(t,e=!0){const i=t;i.addEmitterShapeGenerator&&i.addEmitterShapeGenerator("circle",new ns),await i.refresh(e)}(t,!1),await async function(t,e=!0){const i=t;i.addEmitterShapeGenerator&&i.addEmitterShapeGenerator("square",new cs),await i.refresh(e)}(t,!1),await an(t,e)}return gn(Ti),e})()));
\ No newline at end of file
diff --git a/docs/make.bat b/docs/make.bat
deleted file mode 100644
index 32bb2452..00000000
--- a/docs/make.bat
+++ /dev/null
@@ -1,35 +0,0 @@
-@ECHO OFF
-
-pushd %~dp0
-
-REM Command file for Sphinx documentation
-
-if "%SPHINXBUILD%" == "" (
- set SPHINXBUILD=sphinx-build
-)
-set SOURCEDIR=.
-set BUILDDIR=_build
-
-%SPHINXBUILD% >NUL 2>NUL
-if errorlevel 9009 (
- echo.
- echo.The 'sphinx-build' command was not found. Make sure you have Sphinx
- echo.installed, then set the SPHINXBUILD environment variable to point
- echo.to the full path of the 'sphinx-build' executable. Alternatively you
- echo.may add the Sphinx directory to PATH.
- echo.
- echo.If you don't have Sphinx installed, grab it from
- echo.https://www.sphinx-doc.org/
- exit /b 1
-)
-
-if "%1" == "" goto help
-
-%SPHINXBUILD% -M %1 %SOURCEDIR% %BUILDDIR% %SPHINXOPTS% %O%
-goto end
-
-:help
-%SPHINXBUILD% -M help %SOURCEDIR% %BUILDDIR% %SPHINXOPTS% %O%
-
-:end
-popd
diff --git a/docs/overrides/fancylogo.txt b/docs/overrides/fancylogo.txt
new file mode 100644
index 00000000..230d5476
--- /dev/null
+++ b/docs/overrides/fancylogo.txt
@@ -0,0 +1,257 @@
+---
+hide:
+- navigation
+- toc
+---
+
+
+
+
+
+
+
+
+
+
+
+
+
Loading...
+
+
+
+
+
+
+
+
+
+
\ No newline at end of file
diff --git a/docs/overrides/main.html b/docs/overrides/main.html
new file mode 100644
index 00000000..9114c614
--- /dev/null
+++ b/docs/overrides/main.html
@@ -0,0 +1,12 @@
+{% extends "base.html" %}
+
+{% block content %}
+{% if page.nb_url %}
+
+ {% include ".icons/material/download.svg" %}
+
+
+{% endif %}
+
+{{ super() }}
+{% endblock content %}
diff --git a/docs/requirements.txt b/docs/requirements.txt
deleted file mode 100644
index 627184b2..00000000
--- a/docs/requirements.txt
+++ /dev/null
@@ -1,22 +0,0 @@
-# Install with local version. Need to run from root of repo
-.[graph]
-
-sphinx >6.0, <7.0
-myst-parser ==2.0.0
-nbsphinx >=0.8.5, <=0.8.9
-pandoc >=1.0, <=2.3
-docutils >=0.16, <0.20
-sphinxcontrib-fulltoc >=1.0, <=1.2.0
-sphinxcontrib-mockautodoc
-sphinx-autobuild
-sphinx-autodoc-typehints >=1.16
-sphinx-paramlinks >=0.5.1, <=0.5.4
-sphinx-togglebutton >=0.2, <=0.3.2
-sphinx-copybutton >=0.3, <=0.5.2
-sphinx-multiproject
-sphinx-toolbox ==3.4.0
-sphinx-rtd-dark-mode
-sphinxcontrib-video ==0.2.0
-jinja2 >=3.0.0,<3.2.0
-sphinxcontrib.katex==0.8.6
-sphinx-collections==0.0.1
\ No newline at end of file
diff --git a/docs/stylesheets/extra.css b/docs/stylesheets/extra.css
new file mode 100644
index 00000000..377d9c59
--- /dev/null
+++ b/docs/stylesheets/extra.css
@@ -0,0 +1,32 @@
+/* Custom colors */
+:root {
+ --md-primary-fg-color: #B92B0F;
+ --md-primary-fg-color--light: #F05F42;
+ --md-primary-fg-color--dark: #B92B0F;
+
+ --md-accent-fg-color: #B92B0F;
+ --md-accent-fg-color--transparent: #B92B0F;
+ --md-accent-bg-color: #ffffff;
+ --md-accent-bg-color--light: #B92B0F;
+ }
+
+[data-md-color-scheme="aiforco"] {
+ --md-primary-fg-color: #B92B0F;
+ --md-primary-fg-color--light: #F05F42;
+ --md-primary-fg-color--dark: #B92B0F;
+}
+
+/* [data-md-color-accent=indigo] {
+ --md-accent-fg-color: #B92B0F;
+} */
+
+/* Ensure code blocks wrap text */
+.codehilite pre {
+ white-space: pre-wrap; /* Allow text to wrap within the pre element */
+ word-break: break-word; /* Break the word at the edge of the container if necessary */
+}
+
+/* Improve overall readability of code by adding some padding */
+.codehilite {
+ padding: 8px; /* Adjust padding to fit your design */
+}
diff --git a/docs/stylesheets/mkdocstrings.css b/docs/stylesheets/mkdocstrings.css
new file mode 100644
index 00000000..abea38a6
--- /dev/null
+++ b/docs/stylesheets/mkdocstrings.css
@@ -0,0 +1,54 @@
+/* Indentation. */
+div.doc-contents:not(.first) {
+ padding-left: 15px;
+ border-left: .05rem solid var(--md-typeset-table-color);
+}
+
+
+/* Fancier color for operators such as * and |. */
+.doc-signature .o {
+ color: var(--md-code-hl-special-color);
+}
+
+/* Fancier color for constants such as None, True, and False. */
+.doc-signature .kc {
+ color: var(--md-code-hl-constant-color);
+}
+
+/* Fancier color for built-in types (only useful when cross-references are used). */
+.doc-signature .n > a[href^="https://docs.python.org/"][href*="/functions.html#"],
+.doc-signature .n > a[href^="https://docs.python.org/"][href*="/stdtypes.html#"] {
+ color: var(--md-code-hl-constant-color);
+}
+
+
+/* Nice names only in TOC */
+.doc-symbol-toc.doc-symbol-method::after {
+ content: "m";
+}
+
+.doc-symbol-toc.doc-symbol-function::after {
+ content: "f";
+}
+
+.doc-symbol-toc.doc-symbol-class::after {
+ content: "C";
+}
+
+.doc-symbol-toc.doc-symbol-module::after {
+ content: "M";
+}
+
+.doc-symbol-toc.doc-symbol-attribute::after {
+ content: "A";
+}
+
+.doc-symbol-toc.doc-symbol-parameter::after {
+ content: "P";
+}
+
+/* Line under link as solid */
+.doc-signature .autorefs {
+ color: inherit;
+ border-bottom: 1px solid currentcolor;
+}
\ No newline at end of file
diff --git a/examples/2d-meta_train.py b/examples/2d-meta_train.py
new file mode 100644
index 00000000..1f3fb8d4
--- /dev/null
+++ b/examples/2d-meta_train.py
@@ -0,0 +1,80 @@
+from lightning.pytorch.callbacks import ModelCheckpoint, RichModelSummary
+from lightning.pytorch.loggers import WandbLogger
+
+from rl4co.envs import CVRPEnv
+from rl4co.models.zoo.am import AttentionModelPolicy
+from rl4co.models.zoo.pomo import POMO
+from rl4co.utils.trainer import RL4COTrainer
+from rl4co.utils.meta_trainer import ReptileCallback
+
+def main():
+ # Set device
+ device_id = 0
+
+ # RL4CO env based on TorchRL
+ env = CVRPEnv(generator_params={'num_loc': 50})
+
+ # Policy: neural network, in this case with encoder-decoder architecture
+ # Note that this is adapted the same as POMO did in the original paper
+ policy = AttentionModelPolicy(env_name=env.name,
+ embed_dim=128,
+ num_encoder_layers=6,
+ num_heads=8,
+ normalization="instance",
+ use_graph_context=False
+ )
+
+ # RL Model (POMO)
+ model = POMO(env,
+ policy,
+ batch_size=64, # meta_batch_size
+ train_data_size=64 * 50, # equals to (meta_batch_size) * (gradient decent steps in the inner-loop optimization of meta-learning method)
+ val_data_size=0,
+ optimizer_kwargs={"lr": 1e-4, "weight_decay": 1e-6},
+ )
+
+ # Example callbacks
+ checkpoint_callback = ModelCheckpoint(
+ dirpath="meta_pomo/checkpoints", # save to checkpoints/
+ filename="epoch_{epoch:03d}", # save as epoch_XXX.ckpt
+ save_top_k=1, # save only the best model
+ save_last=True, # save the last model
+ monitor="val/reward", # monitor validation reward
+ mode="max", # maximize validation reward
+ )
+ rich_model_summary = RichModelSummary(max_depth=3) # model summary callback
+
+ # Meta callbacks
+ meta_callback = ReptileCallback(
+ num_tasks = 1, # the number of tasks in a mini-batch, i.e. `B` in the original paper
+ alpha = 0.9, # initial weight of the task model for the outer-loop optimization of reptile
+ alpha_decay = 1, # weight decay of the task model for the outer-loop optimization of reptile. No decay performs better.
+ min_size = 20, # minimum of sampled size in meta tasks (only supported in cross-size generalization)
+ max_size= 150, # maximum of sampled size in meta tasks (only supported in cross-size generalization)
+ data_type="size_distribution", # choose from ["size", "distribution", "size_distribution"]
+ sch_bar=0.9, # for the task scheduler of size setting, where lr_decay_epoch = sch_bar * epochs, i.e. after this epoch, learning rate will decay with a weight 0.1
+ print_log=True # whether to print the sampled tasks in each meta iteration
+ )
+ callbacks = [meta_callback, checkpoint_callback, rich_model_summary]
+
+ # Logger
+ logger = WandbLogger(project="rl4co", name=f"{env.name}_pomo_reptile")
+ # logger = None # uncomment this line if you don't want logging
+
+ # Adjust your trainer to the number of epochs you want to run
+ trainer = RL4COTrainer(
+ max_epochs=15000, # (the number of meta_model updates) * (the number of tasks in a mini-batch)
+ callbacks=callbacks,
+ accelerator="gpu",
+ devices=[device_id],
+ logger=logger,
+ limit_train_batches=50 # gradient decent steps in the inner-loop optimization of meta-learning method
+ )
+
+ # Fit
+ trainer.fit(model)
+
+
+if __name__ == "__main__":
+ main()
+
diff --git a/examples/other/2-scheduling.ipynb b/examples/other/2-scheduling.ipynb
index 4c4c029e..2fc6856b 100644
--- a/examples/other/2-scheduling.ipynb
+++ b/examples/other/2-scheduling.ipynb
@@ -13,15 +13,27 @@
},
{
"cell_type": "code",
- "execution_count": 2,
+ "execution_count": 1,
"metadata": {},
"outputs": [
{
- "name": "stdout",
+ "name": "stderr",
"output_type": "stream",
"text": [
- "The autoreload extension is already loaded. To reload it, use:\n",
- " %reload_ext autoreload\n"
+ "/home/laurin.luttmann/miniconda3/envs/cuda1203/lib/python3.10/site-packages/lightning_utilities/core/imports.py:14: DeprecationWarning: pkg_resources is deprecated as an API. See https://setuptools.pypa.io/en/latest/pkg_resources.html\n",
+ " import pkg_resources\n",
+ "/home/laurin.luttmann/miniconda3/envs/cuda1203/lib/python3.10/site-packages/lightning/fabric/__init__.py:41: Deprecated call to `pkg_resources.declare_namespace('lightning.fabric')`.\n",
+ "Implementing implicit namespace packages (as specified in PEP 420) is preferred to `pkg_resources.declare_namespace`. See https://setuptools.pypa.io/en/latest/references/keywords.html#keyword-namespace-packages\n",
+ "/home/laurin.luttmann/miniconda3/envs/cuda1203/lib/python3.10/site-packages/pkg_resources/__init__.py:2317: DeprecationWarning: Deprecated call to `pkg_resources.declare_namespace('lightning')`.\n",
+ "Implementing implicit namespace packages (as specified in PEP 420) is preferred to `pkg_resources.declare_namespace`. See https://setuptools.pypa.io/en/latest/references/keywords.html#keyword-namespace-packages\n",
+ " declare_namespace(parent)\n",
+ "/home/laurin.luttmann/miniconda3/envs/cuda1203/lib/python3.10/site-packages/lightning/pytorch/__init__.py:37: Deprecated call to `pkg_resources.declare_namespace('lightning.pytorch')`.\n",
+ "Implementing implicit namespace packages (as specified in PEP 420) is preferred to `pkg_resources.declare_namespace`. See https://setuptools.pypa.io/en/latest/references/keywords.html#keyword-namespace-packages\n",
+ "/home/laurin.luttmann/miniconda3/envs/cuda1203/lib/python3.10/site-packages/pkg_resources/__init__.py:2317: DeprecationWarning: Deprecated call to `pkg_resources.declare_namespace('lightning')`.\n",
+ "Implementing implicit namespace packages (as specified in PEP 420) is preferred to `pkg_resources.declare_namespace`. See https://setuptools.pypa.io/en/latest/references/keywords.html#keyword-namespace-packages\n",
+ " declare_namespace(parent)\n",
+ "/home/laurin.luttmann/miniconda3/envs/cuda1203/lib/python3.10/site-packages/tqdm/auto.py:21: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html\n",
+ " from .autonotebook import tqdm as notebook_tqdm\n"
]
}
],
@@ -59,7 +71,7 @@
" \"min_processing_time\": 1, # the minimum time required for a machine to process an operation\n",
" \"max_processing_time\": 20, # the maximum time required for a machine to process an operation\n",
" \"min_eligible_ma_per_op\": 1, # the minimum number of machines capable to process an operation\n",
- " \"max_eligible_ma_per_op\": 3, # the maximum number of machines capable to process an operation\n",
+ " \"max_eligible_ma_per_op\": 2, # the maximum number of machines capable to process an operation\n",
"}"
]
},
@@ -91,11 +103,15 @@
{
"cell_type": "code",
"execution_count": 5,
- "metadata": {},
+ "metadata": {
+ "jupyter": {
+ "source_hidden": true
+ }
+ },
"outputs": [
{
"data": {
- "image/png": "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",
+ "image/png": "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",
"text/plain": [
""
]
@@ -171,7 +187,7 @@
"cell_type": "markdown",
"metadata": {},
"source": [
- "# Build a Model to Solve the FJSP\n",
+ "## Build a Model to Solve the FJSP\n",
"\n",
"In the FJSP we typically encode Operations and Machines separately, since they pose different node types in a k-partite Graph. Therefore, the encoder for the FJSP returns two hidden representations, the first containing machine embeddings and the second containing operation embeddings:"
]
@@ -208,8 +224,8 @@
"name": "stdout",
"output_type": "stream",
"text": [
- "torch.Size([1, 5, 32])\n",
- "torch.Size([1, 60, 32])\n"
+ "torch.Size([1, 60, 32])\n",
+ "torch.Size([1, 5, 32])\n"
]
}
],
@@ -235,7 +251,7 @@
{
"data": {
"text/plain": [
- "tensor([[ 0, 5, 10, 16, 20, 24, 29, 34, 40, 44]])"
+ "tensor([[ 0, 4, 9, 15, 21, 27, 31, 37, 41, 45]])"
]
},
"execution_count": 8,
@@ -250,7 +266,7 @@
},
{
"cell_type": "code",
- "execution_count": 13,
+ "execution_count": 9,
"metadata": {},
"outputs": [
{
@@ -270,7 +286,7 @@
},
{
"cell_type": "code",
- "execution_count": 14,
+ "execution_count": 10,
"metadata": {},
"outputs": [],
"source": [
@@ -286,7 +302,7 @@
"cell_type": "markdown",
"metadata": {},
"source": [
- "# Visualize solution construction\n",
+ "## Visualize solution construction\n",
"\n",
"Starting at $t=0$, the decoder uses the machine-operation embeddings of the encoder to decide which machine-**job**-combination to schedule next. Note, that due to the precedence relationship, the operations to be scheduled next are fixed per job. Therefore, it is sufficient to determine the next job to be scheduled, which significantly reduces the action space. \n",
"\n",
@@ -299,7 +315,7 @@
},
{
"cell_type": "code",
- "execution_count": 15,
+ "execution_count": 20,
"metadata": {},
"outputs": [
{
@@ -313,7 +329,7 @@
},
{
"data": {
- "image/png": "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",
+ "image/png": "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",
"text/plain": [
""
]
@@ -358,97 +374,75 @@
},
{
"cell_type": "code",
- "execution_count": 16,
+ "execution_count": 20,
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "if torch.cuda.is_available():\n",
+ " accelerator = \"gpu\"\n",
+ " batch_size = 256\n",
+ " train_data_size = 2_000\n",
+ " embed_dim = 128\n",
+ " num_encoder_layers = 4\n",
+ "else:\n",
+ " accelerator = \"cpu\"\n",
+ " batch_size = 32\n",
+ " train_data_size = 1_000\n",
+ " embed_dim = 64\n",
+ " num_encoder_layers = 2"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 21,
"metadata": {},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
- "/Users/luttmann/opt/miniconda3/envs/rl4co/lib/python3.9/site-packages/lightning/pytorch/utilities/parsing.py:198: Attribute 'env' is an instance of `nn.Module` and is already saved during checkpointing. It is recommended to ignore them using `self.save_hyperparameters(ignore=['env'])`.\n",
- "/Users/luttmann/opt/miniconda3/envs/rl4co/lib/python3.9/site-packages/lightning/pytorch/utilities/parsing.py:198: Attribute 'policy' is an instance of `nn.Module` and is already saved during checkpointing. It is recommended to ignore them using `self.save_hyperparameters(ignore=['policy'])`.\n",
- "/Users/luttmann/opt/miniconda3/envs/rl4co/lib/python3.9/site-packages/lightning/pytorch/trainer/connectors/accelerator_connector.py:551: You passed `Trainer(accelerator='cpu', precision='16-mixed')` but AMP with fp16 is not supported on CPU. Using `precision='bf16-mixed'` instead.\n",
- "Using bfloat16 Automatic Mixed Precision (AMP)\n",
- "GPU available: False, used: False\n",
+ "Using 16bit Automatic Mixed Precision (AMP)\n",
+ "GPU available: True (cuda), used: True\n",
"TPU available: False, using: 0 TPU cores\n",
"IPU available: False, using: 0 IPUs\n",
"HPU available: False, using: 0 HPUs\n",
- "/Users/luttmann/opt/miniconda3/envs/rl4co/lib/python3.9/site-packages/lightning/pytorch/trainer/connectors/logger_connector/logger_connector.py:67: Starting from v1.9.0, `tensorboardX` has been removed as a dependency of the `lightning.pytorch` package, due to potential conflicts with other packages in the ML ecosystem. For this reason, `logger=True` will use `CSVLogger` as the default logger, unless the `tensorboard` or `tensorboardX` packages are found. Please `pip install lightning[extra]` or one of them to enable TensorBoard support by default\n",
- "Missing logger folder: /Users/luttmann/Documents/Diss/Repos/nco/ai4co/rl4co/examples/other/lightning_logs\n",
"val_file not set. Generating dataset instead\n",
- "test_file not set. Generating dataset instead\n",
- "\n",
- " | Name | Type | Params\n",
- "--------------------------------------------\n",
- "0 | env | FJSPEnv | 0 \n",
- "1 | policy | L2DPolicy | 15.9 K\n",
- "2 | baseline | WarmupBaseline | 15.9 K\n",
- "--------------------------------------------\n",
- "31.9 K Trainable params\n",
- "0 Non-trainable params\n",
- "31.9 K Total params\n",
- "0.127 Total estimated model params size (MB)\n"
- ]
- },
- {
- "data": {
- "application/vnd.jupyter.widget-view+json": {
- "model_id": "c543880423f84865a05170d16a5aa6fd",
- "version_major": 2,
- "version_minor": 0
- },
- "text/plain": [
- "Sanity Checking: | | 0/? [00:00 20\u001b[0m \u001b[43mtrainer\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mfit\u001b[49m\u001b[43m(\u001b[49m\u001b[43mmodel\u001b[49m\u001b[43m)\u001b[49m\n",
+ "File \u001b[0;32m~/repos/ai4co/rl4co/rl4co/utils/trainer.py:146\u001b[0m, in \u001b[0;36mRL4COTrainer.fit\u001b[0;34m(self, model, train_dataloaders, val_dataloaders, datamodule, ckpt_path)\u001b[0m\n\u001b[1;32m 141\u001b[0m log\u001b[38;5;241m.\u001b[39mwarning(\n\u001b[1;32m 142\u001b[0m \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mOverriding gradient_clip_val to None for \u001b[39m\u001b[38;5;124m'\u001b[39m\u001b[38;5;124mautomatic_optimization=False\u001b[39m\u001b[38;5;124m'\u001b[39m\u001b[38;5;124m models\u001b[39m\u001b[38;5;124m\"\u001b[39m\n\u001b[1;32m 143\u001b[0m )\n\u001b[1;32m 144\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mgradient_clip_val \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;01mNone\u001b[39;00m\n\u001b[0;32m--> 146\u001b[0m \u001b[38;5;28;43msuper\u001b[39;49m\u001b[43m(\u001b[49m\u001b[43m)\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mfit\u001b[49m\u001b[43m(\u001b[49m\n\u001b[1;32m 147\u001b[0m \u001b[43m \u001b[49m\u001b[43mmodel\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mmodel\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 148\u001b[0m \u001b[43m \u001b[49m\u001b[43mtrain_dataloaders\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mtrain_dataloaders\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 149\u001b[0m \u001b[43m \u001b[49m\u001b[43mval_dataloaders\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mval_dataloaders\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 150\u001b[0m \u001b[43m \u001b[49m\u001b[43mdatamodule\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mdatamodule\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 151\u001b[0m \u001b[43m \u001b[49m\u001b[43mckpt_path\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mckpt_path\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 152\u001b[0m \u001b[43m\u001b[49m\u001b[43m)\u001b[49m\n",
+ "File \u001b[0;32m~/miniconda3/envs/cuda1203/lib/python3.10/site-packages/lightning/pytorch/trainer/trainer.py:544\u001b[0m, in \u001b[0;36mTrainer.fit\u001b[0;34m(self, model, train_dataloaders, val_dataloaders, datamodule, ckpt_path)\u001b[0m\n\u001b[1;32m 542\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mstate\u001b[38;5;241m.\u001b[39mstatus \u001b[38;5;241m=\u001b[39m TrainerStatus\u001b[38;5;241m.\u001b[39mRUNNING\n\u001b[1;32m 543\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mtraining \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;01mTrue\u001b[39;00m\n\u001b[0;32m--> 544\u001b[0m \u001b[43mcall\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43m_call_and_handle_interrupt\u001b[49m\u001b[43m(\u001b[49m\n\u001b[1;32m 545\u001b[0m \u001b[43m \u001b[49m\u001b[38;5;28;43mself\u001b[39;49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43m_fit_impl\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mmodel\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mtrain_dataloaders\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mval_dataloaders\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mdatamodule\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mckpt_path\u001b[49m\n\u001b[1;32m 546\u001b[0m \u001b[43m\u001b[49m\u001b[43m)\u001b[49m\n",
+ "File \u001b[0;32m~/miniconda3/envs/cuda1203/lib/python3.10/site-packages/lightning/pytorch/trainer/call.py:44\u001b[0m, in \u001b[0;36m_call_and_handle_interrupt\u001b[0;34m(trainer, trainer_fn, *args, **kwargs)\u001b[0m\n\u001b[1;32m 42\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m trainer\u001b[38;5;241m.\u001b[39mstrategy\u001b[38;5;241m.\u001b[39mlauncher \u001b[38;5;129;01mis\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m \u001b[38;5;28;01mNone\u001b[39;00m:\n\u001b[1;32m 43\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m trainer\u001b[38;5;241m.\u001b[39mstrategy\u001b[38;5;241m.\u001b[39mlauncher\u001b[38;5;241m.\u001b[39mlaunch(trainer_fn, \u001b[38;5;241m*\u001b[39margs, trainer\u001b[38;5;241m=\u001b[39mtrainer, \u001b[38;5;241m*\u001b[39m\u001b[38;5;241m*\u001b[39mkwargs)\n\u001b[0;32m---> 44\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43mtrainer_fn\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m 46\u001b[0m \u001b[38;5;28;01mexcept\u001b[39;00m _TunerExitException:\n\u001b[1;32m 47\u001b[0m _call_teardown_hook(trainer)\n",
+ "File \u001b[0;32m~/miniconda3/envs/cuda1203/lib/python3.10/site-packages/lightning/pytorch/trainer/trainer.py:580\u001b[0m, in \u001b[0;36mTrainer._fit_impl\u001b[0;34m(self, model, train_dataloaders, val_dataloaders, datamodule, ckpt_path)\u001b[0m\n\u001b[1;32m 573\u001b[0m \u001b[38;5;28;01massert\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mstate\u001b[38;5;241m.\u001b[39mfn \u001b[38;5;129;01mis\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m \u001b[38;5;28;01mNone\u001b[39;00m\n\u001b[1;32m 574\u001b[0m ckpt_path \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_checkpoint_connector\u001b[38;5;241m.\u001b[39m_select_ckpt_path(\n\u001b[1;32m 575\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mstate\u001b[38;5;241m.\u001b[39mfn,\n\u001b[1;32m 576\u001b[0m ckpt_path,\n\u001b[1;32m 577\u001b[0m model_provided\u001b[38;5;241m=\u001b[39m\u001b[38;5;28;01mTrue\u001b[39;00m,\n\u001b[1;32m 578\u001b[0m model_connected\u001b[38;5;241m=\u001b[39m\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mlightning_module \u001b[38;5;129;01mis\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m \u001b[38;5;28;01mNone\u001b[39;00m,\n\u001b[1;32m 579\u001b[0m )\n\u001b[0;32m--> 580\u001b[0m \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43m_run\u001b[49m\u001b[43m(\u001b[49m\u001b[43mmodel\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mckpt_path\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mckpt_path\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m 582\u001b[0m \u001b[38;5;28;01massert\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mstate\u001b[38;5;241m.\u001b[39mstopped\n\u001b[1;32m 583\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mtraining \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;01mFalse\u001b[39;00m\n",
+ "File \u001b[0;32m~/miniconda3/envs/cuda1203/lib/python3.10/site-packages/lightning/pytorch/trainer/trainer.py:949\u001b[0m, in \u001b[0;36mTrainer._run\u001b[0;34m(self, model, ckpt_path)\u001b[0m\n\u001b[1;32m 946\u001b[0m log\u001b[38;5;241m.\u001b[39mdebug(\u001b[38;5;124mf\u001b[39m\u001b[38;5;124m\"\u001b[39m\u001b[38;5;132;01m{\u001b[39;00m\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m\u001b[38;5;18m__class__\u001b[39m\u001b[38;5;241m.\u001b[39m\u001b[38;5;18m__name__\u001b[39m\u001b[38;5;132;01m}\u001b[39;00m\u001b[38;5;124m: preparing data\u001b[39m\u001b[38;5;124m\"\u001b[39m)\n\u001b[1;32m 947\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_data_connector\u001b[38;5;241m.\u001b[39mprepare_data()\n\u001b[0;32m--> 949\u001b[0m \u001b[43mcall\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43m_call_setup_hook\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;28;43mself\u001b[39;49m\u001b[43m)\u001b[49m \u001b[38;5;66;03m# allow user to set up LightningModule in accelerator environment\u001b[39;00m\n\u001b[1;32m 950\u001b[0m log\u001b[38;5;241m.\u001b[39mdebug(\u001b[38;5;124mf\u001b[39m\u001b[38;5;124m\"\u001b[39m\u001b[38;5;132;01m{\u001b[39;00m\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m\u001b[38;5;18m__class__\u001b[39m\u001b[38;5;241m.\u001b[39m\u001b[38;5;18m__name__\u001b[39m\u001b[38;5;132;01m}\u001b[39;00m\u001b[38;5;124m: configuring model\u001b[39m\u001b[38;5;124m\"\u001b[39m)\n\u001b[1;32m 951\u001b[0m call\u001b[38;5;241m.\u001b[39m_call_configure_model(\u001b[38;5;28mself\u001b[39m)\n",
+ "File \u001b[0;32m~/miniconda3/envs/cuda1203/lib/python3.10/site-packages/lightning/pytorch/trainer/call.py:94\u001b[0m, in \u001b[0;36m_call_setup_hook\u001b[0;34m(trainer)\u001b[0m\n\u001b[1;32m 92\u001b[0m _call_lightning_datamodule_hook(trainer, \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124msetup\u001b[39m\u001b[38;5;124m\"\u001b[39m, stage\u001b[38;5;241m=\u001b[39mfn)\n\u001b[1;32m 93\u001b[0m _call_callback_hooks(trainer, \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124msetup\u001b[39m\u001b[38;5;124m\"\u001b[39m, stage\u001b[38;5;241m=\u001b[39mfn)\n\u001b[0;32m---> 94\u001b[0m \u001b[43m_call_lightning_module_hook\u001b[49m\u001b[43m(\u001b[49m\u001b[43mtrainer\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[38;5;124;43msetup\u001b[39;49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mstage\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mfn\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m 96\u001b[0m trainer\u001b[38;5;241m.\u001b[39mstrategy\u001b[38;5;241m.\u001b[39mbarrier(\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mpost_setup\u001b[39m\u001b[38;5;124m\"\u001b[39m)\n",
+ "File \u001b[0;32m~/miniconda3/envs/cuda1203/lib/python3.10/site-packages/lightning/pytorch/trainer/call.py:157\u001b[0m, in \u001b[0;36m_call_lightning_module_hook\u001b[0;34m(trainer, hook_name, pl_module, *args, **kwargs)\u001b[0m\n\u001b[1;32m 154\u001b[0m pl_module\u001b[38;5;241m.\u001b[39m_current_fx_name \u001b[38;5;241m=\u001b[39m hook_name\n\u001b[1;32m 156\u001b[0m \u001b[38;5;28;01mwith\u001b[39;00m trainer\u001b[38;5;241m.\u001b[39mprofiler\u001b[38;5;241m.\u001b[39mprofile(\u001b[38;5;124mf\u001b[39m\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124m[LightningModule]\u001b[39m\u001b[38;5;132;01m{\u001b[39;00mpl_module\u001b[38;5;241m.\u001b[39m\u001b[38;5;18m__class__\u001b[39m\u001b[38;5;241m.\u001b[39m\u001b[38;5;18m__name__\u001b[39m\u001b[38;5;132;01m}\u001b[39;00m\u001b[38;5;124m.\u001b[39m\u001b[38;5;132;01m{\u001b[39;00mhook_name\u001b[38;5;132;01m}\u001b[39;00m\u001b[38;5;124m\"\u001b[39m):\n\u001b[0;32m--> 157\u001b[0m output \u001b[38;5;241m=\u001b[39m \u001b[43mfn\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m 159\u001b[0m \u001b[38;5;66;03m# restore current_fx when nested context\u001b[39;00m\n\u001b[1;32m 160\u001b[0m pl_module\u001b[38;5;241m.\u001b[39m_current_fx_name \u001b[38;5;241m=\u001b[39m prev_fx_name\n",
+ "File \u001b[0;32m~/repos/ai4co/rl4co/rl4co/models/rl/common/base.py:155\u001b[0m, in \u001b[0;36mRL4COLitModule.setup\u001b[0;34m(self, stage)\u001b[0m\n\u001b[1;32m 153\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mdataloader_names \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;01mNone\u001b[39;00m\n\u001b[1;32m 154\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39msetup_loggers()\n\u001b[0;32m--> 155\u001b[0m \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mpost_setup_hook\u001b[49m\u001b[43m(\u001b[49m\u001b[43m)\u001b[49m\n",
+ "File \u001b[0;32m~/repos/ai4co/rl4co/rl4co/models/rl/reinforce/reinforce.py:119\u001b[0m, in \u001b[0;36mREINFORCE.post_setup_hook\u001b[0;34m(self, stage)\u001b[0m\n\u001b[1;32m 117\u001b[0m \u001b[38;5;28;01mdef\u001b[39;00m \u001b[38;5;21mpost_setup_hook\u001b[39m(\u001b[38;5;28mself\u001b[39m, stage\u001b[38;5;241m=\u001b[39m\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mfit\u001b[39m\u001b[38;5;124m\"\u001b[39m):\n\u001b[1;32m 118\u001b[0m \u001b[38;5;66;03m# Make baseline taking model itself and train_dataloader from model as input\u001b[39;00m\n\u001b[0;32m--> 119\u001b[0m \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mbaseline\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43msetup\u001b[49m\u001b[43m(\u001b[49m\n\u001b[1;32m 120\u001b[0m \u001b[43m \u001b[49m\u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mpolicy\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 121\u001b[0m \u001b[43m \u001b[49m\u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43menv\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 122\u001b[0m \u001b[43m \u001b[49m\u001b[43mbatch_size\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mval_batch_size\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 123\u001b[0m \u001b[43m \u001b[49m\u001b[43mdevice\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mget_lightning_device\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;28;43mself\u001b[39;49m\u001b[43m)\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 124\u001b[0m \u001b[43m \u001b[49m\u001b[43mdataset_size\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mdata_cfg\u001b[49m\u001b[43m[\u001b[49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[38;5;124;43mval_data_size\u001b[39;49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[43m]\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 125\u001b[0m \u001b[43m \u001b[49m\u001b[43m)\u001b[49m\n",
+ "File \u001b[0;32m~/repos/ai4co/rl4co/rl4co/models/rl/reinforce/baselines.py:117\u001b[0m, in \u001b[0;36mWarmupBaseline.setup\u001b[0;34m(self, *args, **kw)\u001b[0m\n\u001b[1;32m 116\u001b[0m \u001b[38;5;28;01mdef\u001b[39;00m \u001b[38;5;21msetup\u001b[39m(\u001b[38;5;28mself\u001b[39m, \u001b[38;5;241m*\u001b[39margs, \u001b[38;5;241m*\u001b[39m\u001b[38;5;241m*\u001b[39mkw):\n\u001b[0;32m--> 117\u001b[0m \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mbaseline\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43msetup\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkw\u001b[49m\u001b[43m)\u001b[49m\n",
+ "File \u001b[0;32m~/repos/ai4co/rl4co/rl4co/models/rl/reinforce/baselines.py:174\u001b[0m, in \u001b[0;36mRolloutBaseline.setup\u001b[0;34m(self, *args, **kw)\u001b[0m\n\u001b[1;32m 173\u001b[0m \u001b[38;5;28;01mdef\u001b[39;00m \u001b[38;5;21msetup\u001b[39m(\u001b[38;5;28mself\u001b[39m, \u001b[38;5;241m*\u001b[39margs, \u001b[38;5;241m*\u001b[39m\u001b[38;5;241m*\u001b[39mkw):\n\u001b[0;32m--> 174\u001b[0m \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43m_update_policy\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkw\u001b[49m\u001b[43m)\u001b[49m\n",
+ "File \u001b[0;32m~/repos/ai4co/rl4co/rl4co/models/rl/reinforce/baselines.py:187\u001b[0m, in \u001b[0;36mRolloutBaseline._update_policy\u001b[0;34m(self, policy, env, batch_size, device, dataset_size, dataset)\u001b[0m\n\u001b[1;32m 183\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mdataset \u001b[38;5;241m=\u001b[39m env\u001b[38;5;241m.\u001b[39mdataset(batch_size\u001b[38;5;241m=\u001b[39m[dataset_size])\n\u001b[1;32m 185\u001b[0m log\u001b[38;5;241m.\u001b[39minfo(\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mEvaluating baseline policy on evaluation dataset\u001b[39m\u001b[38;5;124m\"\u001b[39m)\n\u001b[1;32m 186\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mbl_vals \u001b[38;5;241m=\u001b[39m (\n\u001b[0;32m--> 187\u001b[0m \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mrollout\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mpolicy\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43menv\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mbatch_size\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mdevice\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mdataset\u001b[49m\u001b[43m)\u001b[49m\u001b[38;5;241m.\u001b[39mcpu()\u001b[38;5;241m.\u001b[39mnumpy()\n\u001b[1;32m 188\u001b[0m )\n\u001b[1;32m 189\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mmean \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mbl_vals\u001b[38;5;241m.\u001b[39mmean()\n",
+ "File \u001b[0;32m~/repos/ai4co/rl4co/rl4co/models/rl/reinforce/baselines.py:242\u001b[0m, in \u001b[0;36mRolloutBaseline.rollout\u001b[0;34m(self, policy, env, batch_size, device, dataset)\u001b[0m\n\u001b[1;32m 238\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m policy(batch, env, decode_type\u001b[38;5;241m=\u001b[39m\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mgreedy\u001b[39m\u001b[38;5;124m\"\u001b[39m)[\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mreward\u001b[39m\u001b[38;5;124m\"\u001b[39m]\n\u001b[1;32m 240\u001b[0m dl \u001b[38;5;241m=\u001b[39m DataLoader(dataset, batch_size\u001b[38;5;241m=\u001b[39mbatch_size, collate_fn\u001b[38;5;241m=\u001b[39mdataset\u001b[38;5;241m.\u001b[39mcollate_fn)\n\u001b[0;32m--> 242\u001b[0m rewards \u001b[38;5;241m=\u001b[39m torch\u001b[38;5;241m.\u001b[39mcat([eval_policy(batch) \u001b[38;5;28;01mfor\u001b[39;00m batch \u001b[38;5;129;01min\u001b[39;00m dl], \u001b[38;5;241m0\u001b[39m)\n\u001b[1;32m 243\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m rewards\n",
+ "File \u001b[0;32m~/repos/ai4co/rl4co/rl4co/models/rl/reinforce/baselines.py:242\u001b[0m, in \u001b[0;36m\u001b[0;34m(.0)\u001b[0m\n\u001b[1;32m 238\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m policy(batch, env, decode_type\u001b[38;5;241m=\u001b[39m\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mgreedy\u001b[39m\u001b[38;5;124m\"\u001b[39m)[\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mreward\u001b[39m\u001b[38;5;124m\"\u001b[39m]\n\u001b[1;32m 240\u001b[0m dl \u001b[38;5;241m=\u001b[39m DataLoader(dataset, batch_size\u001b[38;5;241m=\u001b[39mbatch_size, collate_fn\u001b[38;5;241m=\u001b[39mdataset\u001b[38;5;241m.\u001b[39mcollate_fn)\n\u001b[0;32m--> 242\u001b[0m rewards \u001b[38;5;241m=\u001b[39m torch\u001b[38;5;241m.\u001b[39mcat([\u001b[43meval_policy\u001b[49m\u001b[43m(\u001b[49m\u001b[43mbatch\u001b[49m\u001b[43m)\u001b[49m \u001b[38;5;28;01mfor\u001b[39;00m batch \u001b[38;5;129;01min\u001b[39;00m dl], \u001b[38;5;241m0\u001b[39m)\n\u001b[1;32m 243\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m rewards\n",
+ "File \u001b[0;32m~/repos/ai4co/rl4co/rl4co/models/rl/reinforce/baselines.py:238\u001b[0m, in \u001b[0;36mRolloutBaseline.rollout..eval_policy\u001b[0;34m(batch)\u001b[0m\n\u001b[1;32m 236\u001b[0m \u001b[38;5;28;01mwith\u001b[39;00m torch\u001b[38;5;241m.\u001b[39minference_mode():\n\u001b[1;32m 237\u001b[0m batch \u001b[38;5;241m=\u001b[39m env\u001b[38;5;241m.\u001b[39mreset(batch\u001b[38;5;241m.\u001b[39mto(device))\n\u001b[0;32m--> 238\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43mpolicy\u001b[49m\u001b[43m(\u001b[49m\u001b[43mbatch\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43menv\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mdecode_type\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[38;5;124;43mgreedy\u001b[39;49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[43m)\u001b[49m[\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mreward\u001b[39m\u001b[38;5;124m\"\u001b[39m]\n",
+ "File \u001b[0;32m~/miniconda3/envs/cuda1203/lib/python3.10/site-packages/torch/nn/modules/module.py:1532\u001b[0m, in \u001b[0;36mModule._wrapped_call_impl\u001b[0;34m(self, *args, **kwargs)\u001b[0m\n\u001b[1;32m 1530\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_compiled_call_impl(\u001b[38;5;241m*\u001b[39margs, \u001b[38;5;241m*\u001b[39m\u001b[38;5;241m*\u001b[39mkwargs) \u001b[38;5;66;03m# type: ignore[misc]\u001b[39;00m\n\u001b[1;32m 1531\u001b[0m \u001b[38;5;28;01melse\u001b[39;00m:\n\u001b[0;32m-> 1532\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43m_call_impl\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n",
+ "File \u001b[0;32m~/miniconda3/envs/cuda1203/lib/python3.10/site-packages/torch/nn/modules/module.py:1541\u001b[0m, in \u001b[0;36mModule._call_impl\u001b[0;34m(self, *args, **kwargs)\u001b[0m\n\u001b[1;32m 1536\u001b[0m \u001b[38;5;66;03m# If we don't have any hooks, we want to skip the rest of the logic in\u001b[39;00m\n\u001b[1;32m 1537\u001b[0m \u001b[38;5;66;03m# this function, and just call forward.\u001b[39;00m\n\u001b[1;32m 1538\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m (\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_backward_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_backward_pre_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_forward_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_forward_pre_hooks\n\u001b[1;32m 1539\u001b[0m \u001b[38;5;129;01mor\u001b[39;00m _global_backward_pre_hooks \u001b[38;5;129;01mor\u001b[39;00m _global_backward_hooks\n\u001b[1;32m 1540\u001b[0m \u001b[38;5;129;01mor\u001b[39;00m _global_forward_hooks \u001b[38;5;129;01mor\u001b[39;00m _global_forward_pre_hooks):\n\u001b[0;32m-> 1541\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43mforward_call\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m 1543\u001b[0m \u001b[38;5;28;01mtry\u001b[39;00m:\n\u001b[1;32m 1544\u001b[0m result \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;01mNone\u001b[39;00m\n",
+ "File \u001b[0;32m~/repos/ai4co/rl4co/rl4co/models/common/constructive/base.py:231\u001b[0m, in \u001b[0;36mConstructivePolicy.forward\u001b[0;34m(self, td, env, phase, calc_reward, return_actions, return_entropy, return_hidden, return_init_embeds, return_sum_log_likelihood, actions, max_steps, **decoding_kwargs)\u001b[0m\n\u001b[1;32m 229\u001b[0m \u001b[38;5;28;01mwhile\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m td[\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mdone\u001b[39m\u001b[38;5;124m\"\u001b[39m]\u001b[38;5;241m.\u001b[39mall():\n\u001b[1;32m 230\u001b[0m logits, mask \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mdecoder(td, hidden, num_starts)\n\u001b[0;32m--> 231\u001b[0m td \u001b[38;5;241m=\u001b[39m \u001b[43mdecode_strategy\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mstep\u001b[49m\u001b[43m(\u001b[49m\n\u001b[1;32m 232\u001b[0m \u001b[43m \u001b[49m\u001b[43mlogits\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 233\u001b[0m \u001b[43m \u001b[49m\u001b[43mmask\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 234\u001b[0m \u001b[43m \u001b[49m\u001b[43mtd\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 235\u001b[0m \u001b[43m \u001b[49m\u001b[43maction\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mactions\u001b[49m\u001b[43m[\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mstep\u001b[49m\u001b[43m]\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;28;43;01mif\u001b[39;49;00m\u001b[43m \u001b[49m\u001b[43mactions\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;129;43;01mis\u001b[39;49;00m\u001b[43m \u001b[49m\u001b[38;5;129;43;01mnot\u001b[39;49;00m\u001b[43m \u001b[49m\u001b[38;5;28;43;01mNone\u001b[39;49;00m\u001b[43m \u001b[49m\u001b[38;5;28;43;01melse\u001b[39;49;00m\u001b[43m \u001b[49m\u001b[38;5;28;43;01mNone\u001b[39;49;00m\u001b[43m,\u001b[49m\n\u001b[1;32m 236\u001b[0m \u001b[43m \u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m 237\u001b[0m td \u001b[38;5;241m=\u001b[39m env\u001b[38;5;241m.\u001b[39mstep(td)[\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mnext\u001b[39m\u001b[38;5;124m\"\u001b[39m]\n\u001b[1;32m 238\u001b[0m step \u001b[38;5;241m+\u001b[39m\u001b[38;5;241m=\u001b[39m \u001b[38;5;241m1\u001b[39m\n",
+ "File \u001b[0;32m~/repos/ai4co/rl4co/rl4co/utils/decoding.py:343\u001b[0m, in \u001b[0;36mDecodingStrategy.step\u001b[0;34m(self, logits, mask, td, action, **kwargs)\u001b[0m\n\u001b[1;32m 340\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mmask_logits: \u001b[38;5;66;03m# set mask_logit to None if mask_logits is False\u001b[39;00m\n\u001b[1;32m 341\u001b[0m mask \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;01mNone\u001b[39;00m\n\u001b[0;32m--> 343\u001b[0m logprobs \u001b[38;5;241m=\u001b[39m \u001b[43mprocess_logits\u001b[49m\u001b[43m(\u001b[49m\n\u001b[1;32m 344\u001b[0m \u001b[43m \u001b[49m\u001b[43mlogits\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 345\u001b[0m \u001b[43m \u001b[49m\u001b[43mmask\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 346\u001b[0m \u001b[43m \u001b[49m\u001b[43mtemperature\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mtemperature\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 347\u001b[0m \u001b[43m \u001b[49m\u001b[43mtop_p\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mtop_p\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 348\u001b[0m \u001b[43m \u001b[49m\u001b[43mtop_k\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mtop_k\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 349\u001b[0m \u001b[43m \u001b[49m\u001b[43mtanh_clipping\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mtanh_clipping\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 350\u001b[0m \u001b[43m \u001b[49m\u001b[43mmask_logits\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mmask_logits\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 351\u001b[0m \u001b[43m\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m 352\u001b[0m logprobs, selected_action, td \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_step(\n\u001b[1;32m 353\u001b[0m logprobs, mask, td, action\u001b[38;5;241m=\u001b[39maction, \u001b[38;5;241m*\u001b[39m\u001b[38;5;241m*\u001b[39mkwargs\n\u001b[1;32m 354\u001b[0m )\n\u001b[1;32m 356\u001b[0m \u001b[38;5;66;03m# directly return for improvement methods, since the action for improvement methods is finalized in its own policy\u001b[39;00m\n",
+ "File \u001b[0;32m~/repos/ai4co/rl4co/rl4co/utils/decoding.py:177\u001b[0m, in \u001b[0;36mprocess_logits\u001b[0;34m(logits, mask, temperature, top_p, top_k, tanh_clipping, mask_logits)\u001b[0m\n\u001b[1;32m 175\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m mask_logits:\n\u001b[1;32m 176\u001b[0m \u001b[38;5;28;01massert\u001b[39;00m mask \u001b[38;5;129;01mis\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m \u001b[38;5;28;01mNone\u001b[39;00m, \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mmask must be provided if mask_logits is True\u001b[39m\u001b[38;5;124m\"\u001b[39m\n\u001b[0;32m--> 177\u001b[0m \u001b[43mlogits\u001b[49m\u001b[43m[\u001b[49m\u001b[38;5;241;43m~\u001b[39;49m\u001b[43mmask\u001b[49m\u001b[43m]\u001b[49m \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mfloat\u001b[39m(\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124m-inf\u001b[39m\u001b[38;5;124m\"\u001b[39m)\n\u001b[1;32m 179\u001b[0m logits \u001b[38;5;241m=\u001b[39m logits \u001b[38;5;241m/\u001b[39m temperature \u001b[38;5;66;03m# temperature scaling\u001b[39;00m\n\u001b[1;32m 181\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m top_k \u001b[38;5;241m>\u001b[39m \u001b[38;5;241m0\u001b[39m:\n",
+ "\u001b[0;31mIndexError\u001b[0m: The shape of the mask [256, 11] at index 1 does not match the shape of the indexed tensor [256, 101] at index 1"
]
}
],
"source": [
- "if torch.cuda.is_available():\n",
- " accelerator = \"gpu\"\n",
- " batch_size = 512\n",
- " train_data_size = 100_000\n",
- " embed_dim = 128\n",
- " num_encoder_layers = 4\n",
- "else:\n",
- " accelerator = \"cpu\"\n",
- " batch_size = 32\n",
- " train_data_size = 1_000\n",
- " embed_dim = 64\n",
- " num_encoder_layers = 2\n",
- "\n",
"# Policy: neural network, in this case with encoder-decoder architecture\n",
"policy = L2DPolicy(embed_dim=embed_dim, num_encoder_layers=num_encoder_layers, env_name=\"fjsp\")\n",
"\n",
@@ -470,11 +464,207 @@
"\n",
"trainer.fit(model)"
]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "# Solving the Job-Shop Scheduling Problem (JSSP)"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 2,
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "import gc\n",
+ "from rl4co.envs import JSSPEnv\n",
+ "from rl4co.models.zoo.l2d.model import L2DPPOModel\n",
+ "from rl4co.models.zoo.l2d.policy import L2DPolicy4PPO\n",
+ "from torch.utils.data import DataLoader"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 3,
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "# Lets generate a more complex instance\n",
+ "\n",
+ "generator_params = {\n",
+ " \"num_jobs\": 15, # the total number of jobs\n",
+ " \"num_machines\": 15, # the total number of machines that can process operations\n",
+ " \"min_processing_time\": 1, # the minimum time required for a machine to process an operation\n",
+ " \"max_processing_time\": 99, # the maximum time required for a machine to process an operation\n",
+ "}\n",
+ "\n",
+ "env = JSSPEnv(\n",
+ " generator_params=generator_params, \n",
+ " _torchrl_mode=True, \n",
+ " stepwise_reward=True\n",
+ ")"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "## Train on synthetic data and test on Taillard benchmark"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 36,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stderr",
+ "output_type": "stream",
+ "text": [
+ "Using 16bit Automatic Mixed Precision (AMP)\n",
+ "GPU available: True (cuda), used: True\n",
+ "TPU available: False, using: 0 TPU cores\n",
+ "IPU available: False, using: 0 IPUs\n",
+ "HPU available: False, using: 0 HPUs\n",
+ "Overriding gradient_clip_val to None for 'automatic_optimization=False' models\n",
+ "val_file not set. Generating dataset instead\n",
+ "Provided file name data/../../data/jssp/taillard/15j_15m not found. Make sure to provide a file in the right path first or unset test_file to generate data automatically instead\n",
+ "LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0,1,2,3,4]\n",
+ "\n",
+ " | Name | Type | Params\n",
+ "---------------------------------------------\n",
+ "0 | env | JSSPEnv | 0 \n",
+ "1 | policy | L2DPolicy4PPO | 133 K \n",
+ "2 | policy_old | L2DPolicy4PPO | 133 K \n",
+ "---------------------------------------------\n",
+ "266 K Trainable params\n",
+ "0 Non-trainable params\n",
+ "266 K Total params\n",
+ "1.066 Total estimated model params size (MB)\n"
+ ]
+ },
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "Epoch 0: 100%|β| 8/8 [03:40<00:00, 0.04it/s, v_num=9, train/loss=1.45e+3, train\n",
+ "Validation: | | 0/? [00:00AI4CO contributors
+docs_dir: .
+
+nav:
+ - Home: README.md
+ - Getting Started:
+ - Installation: docs/content/start/installation.md
+ - Quick Start: examples/1-quickstart.ipynb
+ - Training with Hydra: docs/content/start/hydra.md
+ - Overview:
+ - docs/content/intro/intro.md
+ - docs/content/intro/environments.md
+ - docs/content/intro/policies.md
+ - docs/content/intro/rl.md
+ - Tutorials:
+ - Main:
+ - Quickstart: examples/1-quickstart.ipynb
+ - Training a Model: examples/2-full-training.ipynb
+ - Solving New Problems: examples/3-creating-new-env-model.ipynb
+ - Modeling:
+ - Decoding Strategies: examples/modeling/1-decoding-strategies.ipynb
+ - Transductive Methods: examples/modeling/2-transductive-methods.ipynb
+ - Encoder Customization: examples/modeling/3-change-encoder.ipynb
+ - Advanced:
+ - Hydra Configuration: examples/advanced/1-hydra-config.ipynb
+ - Local Search: examples/advanced/3-local-search.ipynb
+ - Routing:
+ - Multi-Task VRP: examples/other/1-mtvrp.ipynb
+ - TSPLib: examples/datasets/1-test-on-tsplib.ipynb
+ - CVRPLib: examples/datasets/2-test-on-cvrplib.ipynb
+ - Scheduling:
+ - Flexible Job Shop Scheduling: examples/other/2-scheduling.ipynb
+ - API Reference:
+ - Environments:
+ - Base Classes: docs/content/api/envs/base.md
+ - Routing Problems: docs/content/api/envs/routing.md
+ - Scheduling Problems: docs/content/api/envs/scheduling.md
+ - EDA Problems: docs/content/api/envs/eda.md
+ - Networks:
+ - Policy Base Classes: docs/content/api/networks/base_policies.md
+ - Neural Network Modules: docs/content/api/networks/nn.md
+ - Environment Embeddings: docs/content/api/networks/env_embeddings.md
+ - RL Algorithms:
+ - Base Classes: docs/content/api/rl/base.md
+ - Reinforce: docs/content/api/rl/reinforce.md
+ - PPO: docs/content/api/rl/ppo.md
+ - A2C: docs/content/api/rl/a2c.md
+ - Zoo:
+ - Constructive AR Methods: docs/content/api/zoo/constructive_ar.md
+ - Constructive NAR Methods: docs/content/api/zoo/constructive_nar.md
+ - Improvement Methods: docs/content/api/zoo/improvement.md
+ - Transductive Methods: docs/content/api/zoo/transductive.md
+ - Additional APIs:
+ - Train and Evaluation: docs/content/api/tasks.md
+ - Decoding Strategies: docs/content/api/decoding.md
+ - Data: docs/content/api/data.md
+ - About:
+ - Contributing to RL4CO: docs/content/general/contribute.md
+ - FAQ: docs/content/general/faq.md
+ - Join AI4CO: docs/content/general/ai4co.md
+
+extra:
+ social:
+ - icon: fontawesome/brands/github
+ link: https://github.com/ai4co
+ - icon: fontawesome/brands/slack
+ link: https://bit.ly/ai4co-slack
+ - icon: fontawesome/solid/globe
+ link: https://ai4co.org
+
+theme:
+ name: material
+ logo: docs/assets/figs/rl4co-logo.svg
+ favicon: docs/assets/figs/rl4co-logo.svg
+ custom_dir: docs/overrides
+ primary: red
+ features:
+ - announce.dismiss
+ - content.code.copy
+ - content.code.annotate
+ - content.code.select
+ - content.tabs.link
+ - content.tooltips
+ - navigation.expand
+ - navigation.instant
+ - navigation.instant.prefetch
+ - navigation.instant.preview
+ - navigation.instant.progress
+ - navigation.indexes
+ - navigation.path
+ - navigation.sections
+ - navigation.top
+ - navigation.tracking
+ - navigation.tabs # horizontal tabs for navigation
+ - search.suggest
+ - search.highlight
+ - search.share
+ - toc.follow
+ palette:
+ - media: "(prefers-color-scheme)"
+ toggle:
+ icon: material/brightness-auto
+ name: Switch to light mode
+ - media: "(prefers-color-scheme: light)"
+ scheme: default
+ primary: custom
+ accent: custom
+ toggle:
+ icon: material/weather-sunny
+ name: Switch to dark mode
+ - media: "(prefers-color-scheme: dark)"
+ scheme: slate
+ primary: custom
+ accent: custom
+ toggle:
+ icon: material/weather-night
+ name: Switch to system preference
+ font:
+ text: Noto Sans
+ code: Fira Code
+
+extra_css:
+ - docs/stylesheets/extra.css # Our custom CSS
+ - docs/stylesheets/mkdocstrings.css # mkdocstrings CSS
+ - https://unpkg.com/katex@0/dist/katex.min.css # Latex math rendering
+
+extra_javascript:
+ - docs/js/katex.js # Latex math rendering
+ - docs/js/particles.min.js
+ - https://unpkg.com/katex@0/dist/katex.min.js # Latex math rendering
+ - https://unpkg.com/katex@0/dist/contrib/auto-render.min.js # Latex math rendering
+
+extra:
+ version:
+ default: stable
+
+plugins:
+ - mkdocstrings:
+ enable_inventory: true # create .inv file for auto linking
+ handlers:
+ python:
+ paths: [rl4co]
+ options:
+ members_order: source
+ separate_signature: true
+ filters: ["!^_"]
+ docstring_options:
+ ignore_init_summary: true
+ merge_init_into_class: true
+ show_signature_annotations: true
+ docstring_section_style: list
+ heading_level: 2
+ inherited_members: false
+ show_root_heading: true
+ show_root_toc_entry: false
+ show_root_full_path: false
+ show_source: true
+ show_symbol_type_heading: true
+ show_symbol_type_toc: true
+ signature_crossrefs: true
+ summary: true
+ extensions:
+ - griffe_inherited_docstrings
+ import:
+ - https://docs.python.org/3/objects.inv
+ - https://pytorch.org/rl/stable/objects.inv
+ - https://pytorch.org/tensordict/stable/objects.inv
+ - https://pytorch.org/docs/stable/objects.inv
+ - https://lightning.ai/docs/pytorch/stable/objects.inv
+ - mike:
+ alias_type: symlink
+ canonical_version: latest
+ - mkdocs-jupyter:
+ include_source: True
+ include: ["*ipynb"]
+ - search
+ - autorefs
+ - autolinks
+ - same-dir
+
+hooks:
+ - docs/hooks.py
+
+markdown_extensions:
+ - pymdownx.details
+ - pymdownx.superfences
+ - pymdownx.arithmatex:
+ generic: true
+ - pymdownx.highlight:
+ pygments_lang_class: true
+ - pymdownx.inlinehilite
+ - pymdownx.snippets
+ - pymdownx.magiclink
+ - pymdownx.superfences
+ - tables
+ - toc:
+ permalink: true
+ title: Page contents
+ - admonition
+ - pymdownx.extra
+ - pymdownx.emoji:
+ emoji_index: !!python/name:material.extensions.emoji.twemoji
+ emoji_generator: !!python/name:material.extensions.emoji.to_svg
+ - pymdownx.tabbed:
+ alternate_style: true
+ - attr_list
+ - md_in_html
+ - def_list
+ - mdx_truly_sane_lists
+ - mdx_breakless_lists
+ - gfm_admonition
+
diff --git a/pyproject.toml b/pyproject.toml
index 06bf7b2c..39c55371 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -1,57 +1,47 @@
-[build-system]
-requires = ["setuptools>=61.0"]
-build-backend = "setuptools.build_meta"
-
-[tool.setuptools.packages.find]
-include = ['rl4co*']
-
-# Automatically determine version from package
-[tool.setuptools.dynamic]
-version = {attr = "rl4co.__version__"}
-
-[project]
+[tool.poetry]
name = "rl4co"
-readme = "README.md"
-
-
+version = "0.5.0.dev2"
description = "RL4CO: an Extensive Reinforcement Learning for Combinatorial Optimization Benchmark"
authors = [
- {name = "Federico Berto", email = "berto.federico2@gmail.com"},
- {name = "Chuanbo Hua", email="cbhua@kaist.ac.kr"},
- {name = "Junyoung Park", email="junyoungpark.ml@gmail.com"},
- {name = "Laurin Luttmann", email="laurin.luttmann@gmail.com"},
- {name = "Yining Ma"},
- {name = "Fanchen Bu"},
- {name = "Jiarui Wang"},
- {name = "Haoran Ye"},
- {name = "Minsu Kim"},
- {name = "Sanghyeok Choi"},
- {name = "Zepeda Gast"},
- {name = "Andre Hottung"},
- {name = "Jianan Zhou"},
- {name = "Jieyi Bi"},
- {name = "Yu Hu"},
- {name = "Fei Liu"},
- {name = "Hyeonah Kim"},
- {name = "Jiwoo Son"},
- {name = "Haeyeon Kim"},
- {name = "Davide Angioni"},
- {name = "Wouter Kool"},
- {name = "Zhiguang Cao"},
- {name = "Jie Zhang"},
- {name = "Kijung Shin"},
- {name = "Cathy Wu"},
- {name = "Sungsoo Ahn"},
- {name = "Guojie Song"},
- {name = "Changhyun Kwon"},
- {name = "Lin Xie"},
- {name = "Jinkyoo Park"},
- ]
-dynamic = ["version"]
-
-license = {file = "LICENSE"}
-
-requires-python = ">=3.8"
+ "Federico Berto ",
+ "Chuanbo Hua ",
+ "Junyoung Park ",
+ "Laurin Luttmann ",
+ "Yining Ma",
+ "Fanchen Bu",
+ "Jiarui Wang",
+ "Haoran Ye",
+ "Minsu Kim",
+ "Sanghyeok Choi",
+ "Zepeda Gast",
+ "Andre Hottung",
+ "Jianan Zhou",
+ "Jieyi Bi",
+ "Yu Hu",
+ "Fei Liu",
+ "Hyeonah Kim",
+ "Jiwoo Son",
+ "Haeyeon Kim",
+ "Davide Angioni",
+ "Wouter Kool",
+ "Zhiguang Cao",
+ "Jie Zhang",
+ "Kijung Shin",
+ "Cathy Wu",
+ "Sungsoo Ahn",
+ "Guojie Song",
+ "Changhyun Kwon",
+ "Lin Xie",
+ "Jinkyoo Park",
+ "AI4CO",
+]
+readme = "README.md"
+license = "MIT"
+homepage = "https://rl4.co"
+repository = "https://github.com/ai4co/rl4co"
+documentation = "https://rl4co.readthedocs.io"
+keywords = ["reinforcement learning", "combinatorial optimization", "benchmark"]
+packages = [{ include = "rl4co" }]
classifiers = [
"Programming Language :: Python",
"Programming Language :: Python :: 3",
@@ -59,69 +49,111 @@ classifiers = [
"Programming Language :: Python :: 3.9",
"Programming Language :: Python :: 3.10",
"Programming Language :: Python :: 3.11",
+ "Programming Language :: Python :: 3.12",
"License :: OSI Approved :: MIT License",
"Development Status :: 4 - Beta",
"Intended Audience :: Developers",
"Intended Audience :: Science/Research",
"Operating System :: OS Independent",
- "Topic :: Scientific/Engineering :: Artificial Intelligence",
+ "Topic :: Scientific/Engineering :: Artificial Intelligence"
]
-dependencies = [
- "einops",
- "hydra-core",
- "hydra-colorlog",
- "lightning>=2.1.0",
- "matplotlib",
- "omegaconf",
- "pyrootutils",
- "rich",
- "robust-downloader",
- "scipy",
- "tensordict>=0.2.0",
- "torchrl>=0.2.0",
- "wandb"
-]
-
-[project.optional-dependencies]
+[tool.poetry.urls]
+"Tracker" = "https://github.com/ai4co/rl4co/issues"
+
+[tool.poetry.dependencies]
+# Required dependencies
+python = ">=3.8"
+einops = "*"
+hydra-core = "*"
+hydra-colorlog = "*"
+lightning = ">=2.1.0"
+matplotlib = "*"
+omegaconf = "*"
+pyrootutils = "*"
+rich = "*"
+robust-downloader = "*"
+scipy = "*"
+tensordict = ">=0.4.0"
+torchrl = ">=0.4.0"
+wandb = "*"
+# Dev dependencies
+black = { version = "*", optional = true }
+pre-commit = { version = ">=3.3.3", optional = true }
+ruff = { version = "*", optional = true }
+pytest = { version = "*", optional = true }
+pytest-cov = { version = "*", optional = true }
+# Graph
+torch_geometric = { version = "*", optional = true }
+# Routing
+numba = { version = ">=0.58.1", optional = true }
+pyvrp = { version = ">=0.8.2", optional = true }
+# Docs
+mkdocs = { version = "*", optional = true }
+mkdocs-material = { version = "*", optional = true }
+mkdocstrings-python = { version = "*", optional = true }
+mike = { version = "*", optional = true }
+mkdocs-jupyter = { version = "*", optional = true }
+mkdocs-redirects = { version = "*", optional = true }
+mkdocs-autolinks-plugin = { version = "*", optional = true }
+griffe-typingdoc = { version = "*", optional = true }
+griffe-inherited-docstrings = { version = "*", optional = true }
+griffe = { version = "*", optional = true }
+mkdocs-same-dir = { version = "*", optional = true }
+mdx-breakless-lists = { version = "*", optional = true }
+mdx-truly-sane-lists = { version = "*", optional = true }
+markdown-gfm-admonition = { version = "*", optional = true }
+
+[tool.poetry.extras]
+dev = ["black", "pre-commit", "ruff", "pytest", "pytest-cov"]
graph = ["torch_geometric"]
-testing = ["pytest", "pytest-cov"]
-dev = ["black", "ruff", "pre-commit>=3.3.3"]
-routing = ["numba>=0.58.1", "pyvrp>=0.8.2"]
+routing = ["numba", "pyvrp"]
+docs = [
+ "mkdocs",
+ "mkdocs-material",
+ "mkdocstrings-python",
+ "mike",
+ "mkdocs-jupyter",
+ "mkdocs-redirects",
+ "mkdocs-autolinks-plugin",
+ "griffe-typingdoc",
+ "griffe-inherited-docstrings",
+ "griffe",
+ "black", # for formatting docstrings
+ "mkdocs-same-dir",
+ "mdx-breakless-lists",
+ "mdx-truly-sane-lists",
+ "markdown-gfm-admonition",
+]
-[project.urls]
-"Homepage" = "https://github.com/ai4co/rl4co"
-"Bug Tracker" = "https://github.com/ai4co/rl4co/issues"
+[tool.black]
+line-length = 90
+target-version = ["py311"]
+include = '\.pyi?$'
+exclude = '''
+(
+ /(
+ \.direnv
+ | \.eggs
+ | \.git
+ | \.tox
+ | \.venv
+ | _build
+ | build
+ | dist
+ | venv
+ )/
+)
+'''
-# Ruff + Black + isort combo for code formatting
-# Adapted from https://github.com/AntonOsika/gpt-engineer (kudos!)
-# https://beta.ruff.rs/docs/configuration/#using-pyprojecttoml
[tool.ruff]
select = ["F", "E", "W", "I001"]
line-length = 90
show-fixes = false
-target-version = "py310"
+target-version = "py311"
task-tags = ["TODO", "FIXME"]
-exclude = [
- ".bzr",
- ".direnv",
- ".eggs",
- ".git",
- ".ruff_cache",
- ".svn",
- ".tox",
- ".venv",
- "__pypackages__",
- "_build",
- "buck-out",
- "build",
- "dist",
- "node_modules",
- "venv",
-]
ignore = ["E501"] # never enforce `E501` (line length violations), done in Black
-# Ignore `E402` (import violations) and "F401" (unused imports) in all `__init__.py` files
[tool.ruff.per-file-ignores]
"__init__.py" = ["E402", "F401"]
@@ -139,27 +171,6 @@ combine-as-imports = true
split-on-trailing-comma = false
lines-between-types = 1
-# https://black.readthedocs.io/en/stable/usage_and_configuration/the_basics.html
-[tool.black]
-line-length = 90
-target-version = ["py311"]
-include = '\.pyi?$'
-exclude = '''
-(
- /(
- \.direnv
- | \.eggs
- | \.git
- | \.tox
- | \.venv
- | _build
- | build
- | dist
- | venv
- )/
-)
-'''
-
[tool.coverage]
include = ["rl4co.*"]
@@ -170,3 +181,7 @@ exclude_lines = [
"pragma: no cover",
"if __name__ == .__main__.:",
]
+
+[build-system]
+requires = ["poetry"]
+build-backend = "poetry.core.masonry.api"
\ No newline at end of file
diff --git a/rl4co/__init__.py b/rl4co/__init__.py
index 2d299658..1830875a 100644
--- a/rl4co/__init__.py
+++ b/rl4co/__init__.py
@@ -1 +1,4 @@
-__version__ = "0.5.0dev0"
+from importlib.metadata import version as get_version
+
+# The package version is obtained from the pyproject.toml file
+__version__ = get_version(__package__)
diff --git a/rl4co/data/dataset.py b/rl4co/data/dataset.py
index a1f03391..f3519ada 100644
--- a/rl4co/data/dataset.py
+++ b/rl4co/data/dataset.py
@@ -1,10 +1,44 @@
from typing import Union
+import tensordict
import torch
+from packaging import version
from tensordict.tensordict import TensorDict
from torch.utils.data import Dataset
+# Checks were removed in tensordict 0.5.0, so we should not pass the kwargs
+if version.parse(tensordict.__version__) <= version.parse("0.4.0"):
+ td_kwargs = {"_run_checks": False}
+else:
+ td_kwargs = {}
+
+
+class FastTdDataset(Dataset):
+ """
+ Note:
+ Check out the issue on tensordict for more details:
+ https://github.com/pytorch-labs/tensordict/issues/374.
+ """
+
+ def __init__(self, td: TensorDict):
+ self.data_len = td.batch_size[0]
+ self.data = td
+
+ def __len__(self):
+ return self.data_len
+
+ def __getitems__(self, idx):
+ return self.data[idx]
+
+ def add_key(self, key, value):
+ return ExtraKeyDataset(self, value, key_name=key)
+
+ @staticmethod
+ def collate_fn(batch: Union[dict, TensorDict]):
+ """Collate function compatible with TensorDicts that reassembles a list of dicts."""
+ return batch
+
class TensorDictDataset(Dataset):
"""Dataset compatible with TensorDicts with low CPU usage.
@@ -37,7 +71,7 @@ def collate_fn(batch: Union[dict, TensorDict]):
return TensorDict(
{key: torch.stack([b[key] for b in batch]) for key in batch[0].keys()},
batch_size=torch.Size([len(batch)]),
- _run_checks=False,
+ **td_kwargs,
)
@@ -87,7 +121,7 @@ def __getitems__(self, index):
return TensorDict(
{key: item[index] for key, item in self.data.items()},
batch_size=torch.Size([len(index)]),
- _run_checks=False, # faster this way
+ **td_kwargs,
)
def add_key(self, key, value):
diff --git a/rl4co/envs/__init__.py b/rl4co/envs/__init__.py
index f1de6e68..f3b65b7d 100644
--- a/rl4co/envs/__init__.py
+++ b/rl4co/envs/__init__.py
@@ -28,6 +28,9 @@
from rl4co.envs.scheduling import FFSPEnv, FJSPEnv, SMTWTPEnv
from rl4co.envs.scheduling.jssp.env import JSSPEnv
+# Graph
+from rl4co.envs.graph import MCPEnv, FLPEnv
+
# Register environments
ENV_REGISTRY = {
"atsp": ATSPEnv,
@@ -51,6 +54,8 @@
"mdcpdp": MDCPDPEnv,
"mtvrp": MTVRPEnv,
"tsp_kopt": TSPkoptEnv,
+ "mcp": MCPEnv,
+ "flp": FLPEnv,
}
diff --git a/rl4co/envs/common/base.py b/rl4co/envs/common/base.py
index e477d7e3..d2f4aa55 100644
--- a/rl4co/envs/common/base.py
+++ b/rl4co/envs/common/base.py
@@ -291,7 +291,7 @@ def _set_seed(self, seed: Optional[int]):
self.rng = rng
def to(self, device):
- """Override `to` device method for safety against `None` device (may be found in `TensorDict`))"""
+ """Override `to` device method for safety against `None` device (may be found in `TensorDict`)"""
if device is None:
return self
else:
@@ -336,8 +336,8 @@ def __setstate__(self, state):
class ImprovementEnvBase(RL4COEnvBase, metaclass=abc.ABCMeta):
"""Base class for Improvement environments based on RL4CO EnvBase.
Note that this class assumes that the solution is stored in a linked list format.
- Here, if rec[i] = j, it means the node i is connected to node j, i.e., edge i-j is in the solution.
- For example, if edge 0-1, edge 1-5, edge 2-10 are in the solution, so we have rec[0]=1, rec[1]=5 and rec[2]=10.
+ Here, if `rec[i] = j`, it means the node `i` is connected to node `j`, i.e., edge `i-j` is in the solution.
+ For example, if edge `0-1`, edge `1-5`, edge `2-10` are in the solution, so we have `rec[0]=1`, `rec[1]=5` and `rec[2]=10`.
Kindly see https://github.com/yining043/VRP-DACT/blob/new_version/Play_with_DACT.ipynb for an example at the end for TSP.
"""
diff --git a/rl4co/envs/common/distribution_utils.py b/rl4co/envs/common/distribution_utils.py
new file mode 100644
index 00000000..ab8e1449
--- /dev/null
+++ b/rl4co/envs/common/distribution_utils.py
@@ -0,0 +1,292 @@
+import random
+
+import torch
+
+
+class Cluster:
+ """
+ Multiple gaussian distributed clusters, as in the Solomon benchmark dataset
+ Following the setting in Bi et al. 2022 (https://arxiv.org/abs/2210.07686)
+
+ Args:
+ n_cluster: Number of the gaussian distributed clusters
+ """
+
+ def __init__(self, n_cluster: int = 3):
+ super().__init__()
+ self.lower, self.upper = 0.2, 0.8
+ self.std = 0.07
+ self.n_cluster = n_cluster
+
+ def sample(self, size):
+
+ batch_size, num_loc, _ = size
+
+ # Generate the centers of the clusters
+ center = self.lower + (self.upper - self.lower) * torch.rand(
+ batch_size, self.n_cluster * 2
+ )
+
+ # Pre-define the coordinates
+ coords = torch.zeros(batch_size, num_loc, 2)
+
+ # Calculate the size of each cluster
+ cluster_sizes = [num_loc // self.n_cluster] * self.n_cluster
+ for i in range(num_loc % self.n_cluster):
+ cluster_sizes[i] += 1
+
+ # Generate the coordinates
+ current_index = 0
+ for i in range(self.n_cluster):
+ means = center[:, i * 2 : (i + 1) * 2]
+ stds = torch.full((batch_size, 2), self.std)
+ points = torch.normal(
+ means.unsqueeze(1).expand(-1, cluster_sizes[i], -1),
+ stds.unsqueeze(1).expand(-1, cluster_sizes[i], -1),
+ )
+ coords[:, current_index : current_index + cluster_sizes[i], :] = points
+ current_index += cluster_sizes[i]
+
+ # Confine the coordinates to range [0, 1]
+ coords.clamp_(0, 1)
+
+ return coords
+
+
+class Mixed:
+ """
+ 50% nodes sampled from uniform distribution, 50% nodes sampled from gaussian distribution, as in the Solomon benchmark dataset
+ Following the setting in Bi et al. 2022 (https://arxiv.org/abs/2210.07686)
+
+ Args:
+ n_cluster_mix: Number of the gaussian distributed clusters
+ """
+
+ def __init__(self, n_cluster_mix=1):
+ super().__init__()
+ self.lower, self.upper = 0.2, 0.8
+ self.std = 0.07
+ self.n_cluster_mix = n_cluster_mix
+
+ def sample(self, size):
+
+ batch_size, num_loc, _ = size
+
+ # Generate the centers of the clusters
+ center = self.lower + (self.upper - self.lower) * torch.rand(
+ batch_size, self.n_cluster_mix * 2
+ )
+
+ # Pre-define the coordinates sampled under uniform distribution
+ coords = torch.FloatTensor(batch_size, num_loc, 2).uniform_(0, 1)
+
+ # Sample mutated index (default setting: 50% mutation)
+ mutate_idx = torch.stack(
+ [torch.randperm(num_loc)[: num_loc // 2] for _ in range(batch_size)]
+ )
+
+ # Generate the coordinates
+ segment_size = num_loc // (2 * self.n_cluster_mix)
+ remaining_indices = num_loc // 2 - segment_size * (self.n_cluster_mix - 1)
+ sizes = [segment_size] * (self.n_cluster_mix - 1) + [remaining_indices]
+ for i in range(self.n_cluster_mix):
+ indices = mutate_idx[:, sum(sizes[:i]) : sum(sizes[: i + 1])]
+ means_x = center[:, 2 * i].unsqueeze(1).expand(-1, sizes[i])
+ means_y = center[:, 2 * i + 1].unsqueeze(1).expand(-1, sizes[i])
+ coords.scatter_(
+ 1,
+ indices.unsqueeze(-1).expand(-1, -1, 2),
+ torch.stack(
+ [
+ torch.normal(means_x.expand(-1, sizes[i]), self.std),
+ torch.normal(means_y.expand(-1, sizes[i]), self.std),
+ ],
+ dim=2,
+ ),
+ )
+
+ # Confine the coordinates to range [0, 1]
+ coords.clamp_(0, 1)
+
+ return coords
+
+
+class Gaussian_Mixture:
+ """
+ Following Zhou et al. (2023): https://arxiv.org/abs/2305.19587
+
+ Args:
+ num_modes: the number of clusters/modes in the Gaussian Mixture.
+ cdist: scale of the uniform distribution for center generation.
+ """
+
+ def __init__(self, num_modes: int = 0, cdist: int = 0):
+ super().__init__()
+ self.num_modes = num_modes
+ self.cdist = cdist
+
+ def sample(self, size):
+
+ batch_size, num_loc, _ = size
+
+ if self.num_modes == 0: # (0, 0) - uniform
+ return torch.rand((batch_size, num_loc, 2))
+ elif self.num_modes == 1 and self.cdist == 1: # (1, 1) - gaussian
+ return self.generate_gaussian(batch_size, num_loc)
+ else:
+ res = [self.generate_gaussian_mixture(num_loc) for _ in range(batch_size)]
+ return torch.stack(res)
+
+ def generate_gaussian_mixture(self, num_loc):
+ """Following the setting in Zhang et al. 2022 (https://arxiv.org/abs/2204.03236)"""
+
+ # Randomly decide how many points each mode gets
+ nums = torch.multinomial(
+ input=torch.ones(self.num_modes) / self.num_modes,
+ num_samples=num_loc,
+ replacement=True,
+ )
+
+ # Prepare to collect points
+ coords = torch.empty((0, 2))
+
+ # Generate points for each mode
+ for i in range(self.num_modes):
+ num = (nums == i).sum() # Number of points in this mode
+ if num > 0:
+ center = torch.rand((1, 2)) * self.cdist
+ cov = torch.eye(2) # Covariance matrix
+ nxy = torch.distributions.MultivariateNormal(
+ center.squeeze(), covariance_matrix=cov
+ ).sample((num,))
+ coords = torch.cat((coords, nxy), dim=0)
+
+ return self._global_min_max_scaling(coords)
+
+ def generate_gaussian(self, batch_size, num_loc):
+ """Following the setting in Xin et al. 2022 (https://openreview.net/pdf?id=nJuzV-izmPJ)"""
+
+ # Mean and random covariances
+ mean = torch.full((batch_size, num_loc, 2), 0.5)
+ covs = torch.rand(batch_size) # Random covariances between 0 and 1
+
+ # Generate the coordinates
+ coords = torch.zeros((batch_size, num_loc, 2))
+ for i in range(batch_size):
+ # Construct covariance matrix for each sample
+ cov_matrix = torch.tensor([[1.0, covs[i]], [covs[i], 1.0]])
+ m = torch.distributions.MultivariateNormal(
+ mean[i], covariance_matrix=cov_matrix
+ )
+ coords[i] = m.sample()
+
+ # Shuffle the coordinates
+ indices = torch.randperm(coords.size(0))
+ coords = coords[indices]
+
+ return self._batch_normalize_and_center(coords)
+
+ def _global_min_max_scaling(self, coords):
+
+ # Scale the points to [0, 1] using min-max scaling
+ coords_min = coords.min(0, keepdim=True).values
+ coords_max = coords.max(0, keepdim=True).values
+ coords = (coords - coords_min) / (coords_max - coords_min)
+
+ return coords
+
+ def _batch_normalize_and_center(self, coords):
+ # Step 1: Compute min and max along each batch
+ coords_min = coords.min(dim=1, keepdim=True).values
+ coords_max = coords.max(dim=1, keepdim=True).values
+
+ # Step 2: Normalize coordinates to range [0, 1]
+ coords = (
+ coords - coords_min
+ ) # Broadcasting subtracts min value on each coordinate
+ range_max = (
+ (coords_max - coords_min).max(dim=-1, keepdim=True).values
+ ) # The maximum range among both coordinates
+ coords = coords / range_max # Divide by the max range to normalize
+
+ # Step 3: Center the batch in the middle of the [0, 1] range
+ coords = (
+ coords + (1 - coords.max(dim=1, keepdim=True).values) / 2
+ ) # Centering the batch
+
+ return coords
+
+
+class Mix_Distribution:
+ """
+ Mixture of three exemplar distributions in batch-level, i.e. Uniform, Cluster, Mixed
+ Following the setting in Bi et al. 2022 (https://arxiv.org/abs/2210.07686)
+
+ Args:
+ n_cluster: Number of the gaussian distributed clusters in Cluster distribution
+ n_cluster_mix: Number of the gaussian distributed clusters in Mixed distribution
+ """
+
+ def __init__(self, n_cluster=3, n_cluster_mix=1):
+ super().__init__()
+ self.lower, self.upper = 0.2, 0.8
+ self.std = 0.07
+ self.Mixed = Mixed(n_cluster_mix=n_cluster_mix)
+ self.Cluster = Cluster(n_cluster=n_cluster)
+
+ def sample(self, size):
+
+ batch_size, num_loc, _ = size
+
+ # Pre-define the coordinates sampled under uniform distribution
+ coords = torch.FloatTensor(batch_size, num_loc, 2).uniform_(0, 1)
+
+ # Random sample probability for the distribution of each sample
+ p = torch.rand(batch_size)
+
+ # Mixed
+ mask = p <= 0.33
+ n_mixed = mask.sum().item()
+ if n_mixed > 0:
+ coords[mask] = self.Mixed.sample((n_mixed, num_loc, 2))
+
+ # Cluster
+ mask = (p > 0.33) & (p <= 0.66)
+ n_cluster = mask.sum().item()
+ if n_cluster > 0:
+ coords[mask] = self.Cluster.sample((n_cluster, num_loc, 2))
+
+ # The remaining ones are uniformly distributed
+ return coords
+
+
+class Mix_Multi_Distributions:
+ """
+ Mixture of 11 Gaussian-like distributions in batch-level
+ Following the setting in Zhou et al. (2023): https://arxiv.org/abs/2305.19587
+ """
+
+ def __init__(self):
+ super().__init__()
+ self.dist_set = [(0, 0), (1, 1)] + [
+ (m, c) for m in [3, 5, 7] for c in [10, 30, 50]
+ ]
+
+ def sample(self, size):
+ batch_size, num_loc, _ = size
+ coords = torch.zeros(batch_size, num_loc, 2)
+
+ # Pre-select distributions for the entire batch
+ dists = [random.choice(self.dist_set) for _ in range(batch_size)]
+ unique_dists = list(
+ set(dists)
+ ) # Unique distributions to minimize re-instantiation
+
+ # Instantiate Gaussian_Mixture only once per unique distribution
+ gm_instances = {dist: Gaussian_Mixture(*dist) for dist in unique_dists}
+
+ # Batch process where possible
+ for i, dist in enumerate(dists):
+ coords[i] = gm_instances[dist].sample((1, num_loc, 2)).squeeze(0)
+
+ return coords
diff --git a/rl4co/envs/common/utils.py b/rl4co/envs/common/utils.py
index 5d7612f7..043a2168 100644
--- a/rl4co/envs/common/utils.py
+++ b/rl4co/envs/common/utils.py
@@ -6,7 +6,7 @@
from tensordict.tensordict import TensorDict
from torch.distributions import Exponential, Normal, Poisson, Uniform
-
+from rl4co.envs.common.distribution_utils import Cluster, Mixed, Gaussian_Mixture, Mix_Distribution, Mix_Multi_Distributions
class Generator(metaclass=abc.ABCMeta):
"""Base data generator class, to be called with `env.generator(batch_size)`"""
@@ -43,8 +43,10 @@ def get_sampler(
kwargs: Additional arguments for the distribution
Example:
- >>> sampler_uniform = get_sampler("loc", "uniform", 0, 1)
- >>> sampler_normal = get_sampler("loc", "normal", loc_mean=0.5, loc_std=.2)
+ ```python
+ sampler_uniform = get_sampler("loc", "uniform", 0, 1)
+ sampler_normal = get_sampler("loc", "normal", loc_mean=0.5, loc_std=.2)
+ ```
"""
if isinstance(distribution, (int, float)):
return Uniform(low=distribution, high=distribution)
@@ -76,6 +78,16 @@ def get_sampler(
) # todo: should be also `low, high` and any other corner
elif isinstance(distribution, Callable):
return distribution(**kwargs)
+ elif distribution == "gaussian_mixture":
+ return Gaussian_Mixture(num_modes=kwargs['num_modes'], cdist=kwargs['cdist'])
+ elif distribution == "cluster":
+ return Cluster(kwargs['n_cluster'])
+ elif distribution == "mixed":
+ return Mixed(kwargs['n_cluster_mix'])
+ elif distribution == "mix_distribution":
+ return Mix_Distribution(kwargs['n_cluster'], kwargs['n_cluster_mix'])
+ elif distribution == "mix_multi_distributions":
+ return Mix_Multi_Distributions()
else:
raise ValueError(f"Invalid distribution type of {distribution}")
@@ -87,3 +99,4 @@ def batch_to_scalar(param):
if isinstance(param, torch.Tensor):
return param.item()
return param
+
diff --git a/rl4co/envs/graph/__init__.py b/rl4co/envs/graph/__init__.py
new file mode 100644
index 00000000..d49826b1
--- /dev/null
+++ b/rl4co/envs/graph/__init__.py
@@ -0,0 +1,3 @@
+from rl4co.envs.graph.mcp.env import MCPEnv
+from rl4co.envs.graph.flp.env import FLPEnv
+
diff --git a/rl4co/envs/graph/flp/env.py b/rl4co/envs/graph/flp/env.py
new file mode 100644
index 00000000..aa73b3f9
--- /dev/null
+++ b/rl4co/envs/graph/flp/env.py
@@ -0,0 +1,169 @@
+from typing import Optional
+
+import torch
+
+from tensordict.tensordict import TensorDict
+
+from rl4co.envs.common.base import RL4COEnvBase
+from rl4co.utils.ops import gather_by_index
+from rl4co.utils.pylogger import get_pylogger
+
+from .generator import FLPGenerator
+
+log = get_pylogger(__name__)
+
+
+class FLPEnv(RL4COEnvBase):
+ """Facility Location Problem (FLP) environment
+ At each step, the agent chooses a location. The reward is 0 unless enough number of locations are chosen.
+ The reward is (-) the total distance of each location to its closest chosen location.
+
+ Observations:
+ - the locations
+ - the number of locations to choose
+
+ Constraints:
+ - the given number of locations must be chosen
+
+ Finish condition:
+ - the given number of locations are chosen
+
+ Reward:
+ - (minus) the total distance of each location to its closest chosen location
+
+ Args:
+ generator: FLPGenerator instance as the data generator
+ generator_params: parameters for the generator
+ """
+
+ name = "flp"
+
+ def __init__(
+ self,
+ generator: FLPGenerator = None,
+ generator_params: dict = {},
+ check_solution=False,
+ **kwargs,
+ ):
+ super().__init__(**kwargs)
+ if generator is None:
+ generator = FLPGenerator(**generator_params)
+ self.generator = generator
+ self.check_solution = check_solution
+ self._make_spec(self.generator)
+
+ def _step(self, td: TensorDict) -> TensorDict:
+ # action: [batch_size, 1]; the location to be chosen in each instance
+ selected = td["action"]
+ batch_size = selected.shape[0]
+
+ # Update location selection status
+ chosen = td["chosen"].clone() # (batch_size, n_locations)
+ n_points_ = chosen.shape[-1]
+
+ chosen[torch.arange(batch_size).to(td.device), selected] = True
+
+ # We are done if we choose enough locations
+ done = td["i"] >= (td["to_choose"] - 1)
+
+ # The reward is calculated outside via get_reward for efficiency, so we set it to zero here
+ reward = torch.zeros_like(done)
+
+ # Update distances
+ orig_distances = td["orig_distances"] # (batch_size, n_points, n_points)
+
+ cur_min_dist = (
+ gather_by_index(
+ orig_distances, chosen.nonzero(as_tuple=True)[1].view(batch_size, -1)
+ )
+ .view(batch_size, -1, n_points_)
+ .min(dim=1)
+ .values
+ )
+
+ # We cannot choose the already-chosen locations
+ action_mask = ~chosen
+
+ td.update(
+ {
+ "distances": cur_min_dist, # (batch_size, n_points)
+ # states changed by actions
+ "chosen": chosen, # each entry is binary; 1 iff the corresponding facility is chosen
+ "i": td["i"] + 1, # the number of sets we have chosen
+ "action_mask": action_mask,
+ "reward": reward,
+ "done": done,
+ }
+ )
+ return td
+
+ def _reset(self, td: Optional[TensorDict] = None, batch_size=None) -> TensorDict:
+ self.to(td.device)
+
+ return TensorDict(
+ {
+ # given information
+ "locs": td["locs"], # (batch_size, n_points, dim_loc)
+ "orig_distances": td[
+ "orig_distances"
+ ], # (batch_size, n_points, n_points)
+ "distances": td["distances"], # (batch_size, n_points, n_points)
+ # states changed by actions
+ "chosen": torch.zeros(
+ *td["locs"].shape[:-1], dtype=torch.bool, device=td.device
+ ), # each entry is binary; 1 iff the corresponding facility is chosen
+ "to_choose": td["to_choose"], # the number of sets to choose
+ "i": torch.zeros(
+ *batch_size, dtype=torch.int64, device=td.device
+ ), # the number of sets we have chosen
+ "action_mask": torch.ones(
+ *td["locs"].shape[:-1], dtype=torch.bool, device=td.device
+ ),
+ },
+ batch_size=batch_size,
+ )
+
+ def _make_spec(self, generator: FLPGenerator):
+ # TODO: make spec
+ pass
+
+ def _get_reward(self, td: TensorDict, actions: torch.Tensor) -> torch.Tensor:
+ if self.check_solution:
+ self.check_solution_validity(td, actions)
+
+ # The reward is (minus) the total distance from each location to the closest chosen location
+ chosen = td["chosen"] # (batch_size, n_points)
+ batch_size_ = td["chosen"].shape[0]
+ n_points_ = td["chosen"].shape[-1]
+ orig_distances = td["orig_distances"]
+ cur_min_dist = (
+ gather_by_index(
+ orig_distances, chosen.nonzero(as_tuple=True)[1].view(batch_size_, -1)
+ )
+ .view(batch_size_, -1, n_points_)
+ .min(1)
+ .values.sum(-1)
+ )
+ return -cur_min_dist
+
+ @staticmethod
+ def check_solution_validity(td: TensorDict, actions: torch.Tensor) -> None:
+ # TODO: check solution validity
+ pass
+
+ @staticmethod
+ def local_search(td: TensorDict, actions: torch.Tensor, **kwargs) -> torch.Tensor:
+ # TODO: local search
+ pass
+
+ @staticmethod
+ def get_num_starts(td):
+ return td["action_mask"].shape[-1]
+
+ @staticmethod
+ def select_start_nodes(td, num_starts):
+ num_loc = td["action_mask"].shape[-1]
+ return (
+ torch.arange(num_starts, device=td.device).repeat_interleave(td.shape[0])
+ % num_loc
+ )
diff --git a/rl4co/envs/graph/flp/generator.py b/rl4co/envs/graph/flp/generator.py
new file mode 100644
index 00000000..adbc7de6
--- /dev/null
+++ b/rl4co/envs/graph/flp/generator.py
@@ -0,0 +1,74 @@
+import math
+
+from typing import Callable, Union
+
+import torch
+
+from tensordict.tensordict import TensorDict
+from torch.distributions import Uniform
+
+from rl4co.envs.common.utils import Generator, get_sampler
+from rl4co.utils.ops import get_distance_matrix
+from rl4co.utils.pylogger import get_pylogger
+
+log = get_pylogger(__name__)
+
+
+class FLPGenerator(Generator):
+ """Data generator for the Facility Location Problem (FLP).
+
+ Args:
+ num_loc: number of locations in the FLP
+ min_loc: minimum value for the location coordinates
+ max_loc: maximum value for the location coordinates
+ loc_distribution: distribution for the location coordinates
+
+ Returns:
+ A TensorDict with the following keys:
+ locs [batch_size, num_loc, 2]: locations
+ orig_distances [batch_size, num_loc, num_loc]: original distances between locations
+ distances [batch_size, num_loc]: the current minimum distance rom each location to the chosen locations
+ chosen [batch_size, num_loc]: indicators of chosen locations
+ to_choose [batch_size, 1]: number of locations to choose in the FLP
+ """
+
+ def __init__(
+ self,
+ num_loc: int = 100,
+ min_loc: float = 0.0,
+ max_loc: float = 1.0,
+ loc_distribution: Union[int, float, str, type, Callable] = Uniform,
+ to_choose: int = 10,
+ **kwargs,
+ ):
+ self.num_loc = num_loc
+ self.min_loc = min_loc
+ self.max_loc = max_loc
+ self.to_choose = to_choose
+
+ # Location distribution
+ if kwargs.get("loc_sampler", None) is not None:
+ self.loc_sampler = kwargs["loc_sampler"]
+ else:
+ self.loc_sampler = get_sampler(
+ "loc", loc_distribution, min_loc, max_loc, **kwargs
+ )
+
+ def _generate(self, batch_size) -> TensorDict:
+ # Sample locations
+ locs = self.loc_sampler.sample((*batch_size, self.num_loc, 2))
+ distances = get_distance_matrix(locs)
+ max_dist = math.sqrt(2) * (self.max_loc - self.min_loc)
+
+ return TensorDict(
+ {
+ "locs": locs,
+ "orig_distances": distances,
+ "distances": torch.full(
+ (*batch_size, self.num_loc), max_dist, dtype=torch.float
+ ),
+ "chosen": torch.zeros(*batch_size, self.num_loc, dtype=torch.bool),
+ "to_choose": torch.ones(*batch_size, dtype=torch.long) * self.to_choose,
+ },
+ batch_size=batch_size,
+ )
diff --git a/rl4co/envs/graph/mcp/env.py b/rl4co/envs/graph/mcp/env.py
new file mode 100644
index 00000000..3f0275e0
--- /dev/null
+++ b/rl4co/envs/graph/mcp/env.py
@@ -0,0 +1,193 @@
+from typing import Optional
+
+import torch
+
+from tensordict.tensordict import TensorDict
+
+from rl4co.envs.common.base import RL4COEnvBase
+from rl4co.utils.pylogger import get_pylogger
+
+from .generator import MCPGenerator
+
+log = get_pylogger(__name__)
+
+
+class MCPEnv(RL4COEnvBase):
+ """Maximum Coverage Problem (MCP) environment
+ At each step, the agent chooses a set. The reward is 0 unless enough number of sets are chosen.
+ The reward is the total weights of the covered items (i.e., items in any chosen set).
+
+ Observations:
+ - the weights of items
+ - the membership of items in sets
+ - the number of sets to choose
+
+ Constraints:
+ - the given number of sets must be chosen
+
+ Finish condition:
+ - the given number of sets are chosen
+
+ Reward:
+ - the total weights of the covered items (i.e., items in any chosen set)
+
+ Args:
+ generator: MCPGenerator instance as the data generator
+ generator_params: parameters for the generator
+ """
+
+ name = "mcp"
+
+ def __init__(
+ self,
+ generator: MCPGenerator = None,
+ generator_params: dict = {},
+ check_solution=False,
+ **kwargs,
+ ):
+ super().__init__(**kwargs)
+ if generator is None:
+ generator = MCPGenerator(**generator_params)
+ self.generator = generator
+ self.check_solution = check_solution
+ self._make_spec(self.generator)
+
+ def _step(self, td: TensorDict) -> TensorDict:
+ # action: [batch_size, 1]; the set to be chosen in each instance
+ batch_size = td["action"].shape[0]
+ selected = td["action"]
+
+ # Update set selection status
+ chosen = td["chosen"].clone() # (batch_size, n_sets)
+ chosen[torch.arange(batch_size).to(td.device), selected] = True
+
+ # We are done if we choose enough sets
+ done = td["i"] >= (td["n_sets_to_choose"] - 1)
+
+ # The reward is calculated outside via get_reward for efficiency, so we set it to -inf here
+ reward = torch.ones_like(done) * float("-inf")
+
+ remaining_sets = ~chosen # (batch_size, n_sets)
+
+ chosen_membership = chosen.unsqueeze(-1) * td["membership"]
+ chosen_membership_nonzero = chosen_membership.nonzero()
+ remaining_membership = remaining_sets.unsqueeze(-1) * td["membership"]
+
+ batch_indices, set_indices, item_indices = chosen_membership_nonzero.T
+ chosen_items_indices = chosen_membership[
+ batch_indices, set_indices, item_indices
+ ].long()
+
+ batch_size, n_items = td["weights"].shape
+
+ # We have batch_indices and chosen_items_indices
+ # chosen_items: (batch_size, n_items)
+ # for each i, chosen_items[batch_size[i], chosen_items_indices[i]] += 1
+ chosen_items = torch.zeros(batch_size, n_items + 1, device=td.device)
+ chosen_items[batch_indices, chosen_items_indices] += 1
+ chosen_items = chosen_items[:, 1:] # Remove the first column (invalid zeros)
+
+ # chosen_item[i, j] > 0 means item j is chosen in batch i
+ covered_items = (chosen_items > 0).float() # (batch_size, n_items)
+ remaining_items = 1.0 - covered_items # (batch_size, n_items)
+
+ # We cannot choose the already-chosen sets
+ action_mask = ~chosen
+
+ td.update(
+ {
+ "membership": remaining_membership, # (batch_size, n_sets, max_size)
+ "weights": td["weights"] * remaining_items, # (batch_size, n_items)
+ "chosen": chosen,
+ "i": td["i"] + 1,
+ "action_mask": action_mask,
+ "reward": reward,
+ "done": done,
+ }
+ )
+ return td
+
+ def _reset(self, td: Optional[TensorDict] = None, batch_size=None) -> TensorDict:
+ self.to(td.device)
+
+ return TensorDict(
+ {
+ # given information; constant for each given instance
+ "orig_membership": td["membership"], # (batch_size, n_sets, max_size)
+ "membership": td["membership"], # (batch_size, n_sets, max_size)
+ "orig_weights": td["weights"], # (batch_size, n_items)
+ "weights": td["weights"], # (batch_size, n_items)
+ "n_sets_to_choose": td["n_sets_to_choose"], # (batch_size, 1)
+ # states changed by actions
+ "chosen": torch.zeros(
+ *td["membership"].shape[:-1], dtype=torch.bool, device=td.device
+ ), # each entry is binary; 1 iff the corresponding set is chosen
+ "i": torch.zeros(
+ *batch_size, dtype=torch.int64, device=td.device
+ ), # the number of sets we have chosen
+ "action_mask": torch.ones(
+ *td["membership"].shape[:-1], dtype=torch.bool, device=td.device
+ ),
+ },
+ batch_size=batch_size,
+ )
+
+ def _make_spec(self, generator: MCPGenerator):
+ # TODO: make spec
+ pass
+
+ def _get_reward(self, td: TensorDict, actions: torch.Tensor) -> torch.Tensor:
+ if self.check_solution:
+ self.check_solution_validity(td, actions)
+
+ membership = td[
+ "orig_membership"
+ ] # (batch_size, n_sets, max_size); membership[i, j] = the items in set j in batch i (with 0 padding)
+ weights = td["orig_weights"] # (batch_size, n_items)
+ chosen_sets = td["chosen"] # (batch_size, n_set); 1 if chosen, 0 otherwise
+
+ chosen_membership = chosen_sets.unsqueeze(-1) * membership
+ chosen_membership_nonzero = chosen_membership.nonzero()
+
+ batch_indices, set_indices, item_indices = chosen_membership_nonzero.T
+ chosen_items_indices = chosen_membership[
+ batch_indices, set_indices, item_indices
+ ].long()
+
+ batch_size, n_items = weights.shape
+
+ # We have batch_indices and chosen_items_indices
+ # chosen_items: (batch_size, n_items)
+ # For each i, chosen_items[batch_size[i], chosen_items_indices[i]] += 1
+ chosen_items = torch.zeros(batch_size, n_items + 1, device=td.device)
+ chosen_items[batch_indices, chosen_items_indices] += 1
+ chosen_items = chosen_items[:, 1:] # remove the first column
+
+ # chosen_item[i, j] > 0 means item j is chosen in batch i
+ chosen_items = (chosen_items > 0).float()
+ # Compute the total weights of chosen items
+ chosen_weights = torch.sum(chosen_items * weights, dim=-1)
+
+ return chosen_weights
+
+ @staticmethod
+ def check_solution_validity(td: TensorDict, actions: torch.Tensor) -> None:
+ # TODO: check solution validity
+ pass
+
+ @staticmethod
+ def local_search(td: TensorDict, actions: torch.Tensor, **kwargs) -> torch.Tensor:
+ # TODO: local search
+ pass
+
+ @staticmethod
+ def get_num_starts(td):
+ return td["action_mask"].shape[-1]
+
+ @staticmethod
+ def select_start_nodes(td, num_starts):
+ num_sets = td["action_mask"].shape[-1]
+ return (
+ torch.arange(num_starts, device=td.device).repeat_interleave(td.shape[0])
+ % num_sets
+ )
diff --git a/rl4co/envs/graph/mcp/generator.py b/rl4co/envs/graph/mcp/generator.py
new file mode 100644
index 00000000..92b59410
--- /dev/null
+++ b/rl4co/envs/graph/mcp/generator.py
@@ -0,0 +1,111 @@
+import torch
+
+from tensordict.tensordict import TensorDict
+
+from rl4co.envs.common.utils import Generator
+from rl4co.utils.pylogger import get_pylogger
+
+log = get_pylogger(__name__)
+
+
+def remove_repeat(x: torch.Tensor) -> torch.Tensor:
+ """
+ Remove the repeated elements in each row (i.e., the last dimension) of the input tensor x,
+ and change the repeated elements to 0
+
+ Ref: https://stackoverflow.com/questions/62300404
+
+ Args:
+ x: input tensor
+ """
+
+ # sorting the rows so that duplicate values appear together
+ # e.g., first row: [1, 2, 3, 3, 3, 4, 4]
+ y, indices = x.sort(dim=-1)
+
+ # subtracting, so duplicate values will become 0
+ # e.g., first row: [1, 2, 3, 0, 0, 4, 0]
+ y[..., 1:] *= ((y[..., 1:] - y[..., :-1]) != 0).long()
+
+ # retrieving the original indices of elements
+ indices = indices.sort(dim=-1)[1]
+
+ # re-organizing the rows following original order
+ # e.g., first row: [1, 2, 3, 4, 0, 0, 0]
+ return torch.gather(y, -1, indices)
+
+
+class MCPGenerator(Generator):
+ """Data generator for the Maximum Coverage Problem (MCP).
+
+ Args:
+ num_items: number of items in the MCP
+ num_sets: number of sets in the MCP
+ min_weight: minimum value for the item weights
+ max_weight: maximum value for the item weights
+ min_size: minimum size for the sets
+ max_size: maximum size for the sets
+ n_sets_to_choose: number of sets to choose in the MCP
+
+ Returns:
+ A TensorDict with the following keys:
+ membership [batch_size, num_sets, max_size]: membership of items in sets
+ weights [batch_size, num_items]: weights of the items
+ n_sets_to_choose [batch_size, 1]: number of sets to choose in the MCP
+ """
+
+ def __init__(
+ self,
+ num_items: int = 200,
+ num_sets: int = 100,
+ min_weight: int = 1,
+ max_weight: int = 10,
+ min_size: int = 5,
+ max_size: int = 15,
+ n_sets_to_choose: int = 10,
+ **kwargs,
+ ):
+ self.num_items = num_items
+ self.num_sets = num_sets
+ self.min_weight = min_weight
+ self.max_weight = max_weight
+ self.min_size = min_size
+ self.max_size = max_size
+ self.n_sets_to_choose = n_sets_to_choose
+
+ def _generate(self, batch_size) -> TensorDict:
+ try:
+ batch_size = batch_size[0]
+ except TypeError:
+ batch_size = batch_size
+
+ weights_tensor = torch.randint(
+ self.min_weight, self.max_weight + 1, (batch_size, self.num_items)
+ ).float()
+ # Create membership tensor
+ membership_tensor_max_size = torch.randint(
+ 1, self.num_items + 1, (batch_size, self.num_sets, self.max_size)
+ )
+ # Cutoffs for each set
+ cutoffs = torch.randint(
+ self.min_size, self.max_size + 1, (batch_size, self.num_sets)
+ )
+
+ cutoffs_masks = torch.arange(self.max_size).view(1, 1, -1) < cutoffs.unsqueeze(-1)
+ # Take the masked elements, 0 means the item is invalid
+ membership_tensor = (
+ membership_tensor_max_size * cutoffs_masks
+ ) # (batch_size, num_sets, max_size)
+
+ # Remove repeated items in each set
+ membership_tensor = remove_repeat(membership_tensor)
+
+ return TensorDict(
+ {
+ "membership": membership_tensor.float(), # (batch_size, num_sets, max_size)
+ "weights": weights_tensor.float(), # (batch_size, num_items)
+ "n_sets_to_choose": torch.ones(batch_size, 1)
+ * self.n_sets_to_choose, # (batch_size, 1)
+ },
+ batch_size=batch_size,
+ )
diff --git a/rl4co/envs/routing/mtvrp/env.py b/rl4co/envs/routing/mtvrp/env.py
index 9a23895d..c4a32cc3 100644
--- a/rl4co/envs/routing/mtvrp/env.py
+++ b/rl4co/envs/routing/mtvrp/env.py
@@ -26,65 +26,48 @@ class MTVRPEnv(RL4COEnvBase):
Features:
- *Capacity (C)*
- - Each vehicle has a maximum capacity :math:`Q`, restricting the total load that can be in the vehicle at any point of the route.
+ - Each vehicle has a maximum capacity $Q$, restricting the total load that can be in the vehicle at any point of the route.
- The route must be planned such that the sum of demands and pickups for all customers visited does not exceed this capacity.
- *Time Windows (TW)*
- - Every node :math:`i` has an associated time window :math:`[e_i, l_i]` during which service must commence.
- - Additionally, each node has a service time :math:`s_i`. Vehicles must reach node :math:`i` within its time window; early arrivals must wait at the node location until time :math:`e_i`.
+ - Every node $i$ has an associated time window $[e_i, l_i]$ during which service must commence.
+ - Additionally, each node has a service time $s_i$. Vehicles must reach node $i$ within its time window; early arrivals must wait at the node location until time $e_i$.
- *Open Routes (O)*
- Vehicles are not required to return to the depot after serving all customers.
- - Note that this does not need to be counted as a constraint since it can be modelled by setting zero costs on arcs returning to the depot :math:`c_{i0} = 0` from any customer :math:`i \in C`, and not counting the return arc as part of the route.
+ - Note that this does not need to be counted as a constraint since it can be modelled by setting zero costs on arcs returning to the depot $c_{i0} = 0$ from any customer $i \in C$, and not counting the return arc as part of the route.
- *Backhauls (B)*
- Backhauls generalize demand to also account for return shipments. Customers are either linehaul or backhaul customers.
- - Linehaul customers require delivery of a demand :math:`q_i > 0` that needs to be transported from the depot to the customer, whereas backhaul customers need a pickup of an amount :math:`p_i > 0` that is transported from the client back to the depot.
+ - Linehaul customers require delivery of a demand $q_i > 0$ that needs to be transported from the depot to the customer, whereas backhaul customers need a pickup of an amount $p_i > 0$ that is transported from the client back to the depot.
- It is possible for vehicles to serve a combination of linehaul and backhaul customers in a single route, but then any linehaul customers must precede the backhaul customers in the route.
- *Duration Limits (L)*
- Imposes a limit on the total travel duration (or length) of each route, ensuring a balanced workload across vehicles.
The environment covers the following 16 variants depending on the data generation:
-
- +--------------++--------------+----------------+--------------+--------------------+------------------+
- | VRP Variant || Capacity (C) | Open Route (O) | Backhaul (B) | Duration Limit (L) | Time Window (TW) |
- +==============++==============+================+==============+====================+==================+
- | CVRP || β | | | | |
- +--------------++--------------+----------------+--------------+--------------------+------------------+
- | OVRP || β | β | | | |
- +--------------++--------------+----------------+--------------+--------------------+------------------+
- | VRPB || β | | β | | |
- +--------------++--------------+----------------+--------------+--------------------+------------------+
- | VRPL || β | | | β | |
- +--------------++--------------+----------------+--------------+--------------------+------------------+
- | VRPTW || β | | | | β |
- +--------------++--------------+----------------+--------------+--------------------+------------------+
- | OVRPTW || β | β | | | β |
- +--------------++--------------+----------------+--------------+--------------------+------------------+
- | OVRPB || β | β | β | | |
- +--------------++--------------+----------------+--------------+--------------------+------------------+
- | OVRPL || β | β | | β | |
- +--------------++--------------+----------------+--------------+--------------------+------------------+
- | VRPBL || β | | β | β | |
- +--------------++--------------+----------------+--------------+--------------------+------------------+
- | VRPBTW || β | | β | | β |
- +--------------++--------------+----------------+--------------+--------------------+------------------+
- | VRPLTW || β | | | β | β |
- +--------------++--------------+----------------+--------------+--------------------+------------------+
- | OVRPBL || β | β | β | β | |
- +--------------++--------------+----------------+--------------+--------------------+------------------+
- | OVRPBTW || β | β | β | | β |
- +--------------++--------------+----------------+--------------+--------------------+------------------+
- | OVRPLTW || β | β | | β | β |
- +--------------++--------------+----------------+--------------+--------------------+------------------+
- | VRPBLTW || β | | β | β | β |
- +--------------++--------------+----------------+--------------+--------------------+------------------+
- | OVRPBLTW || β | β | β | β | β |
- +--------------++--------------+----------------+--------------+--------------------+------------------+
+
+ | VRP Variant | Capacity (C) | Open Route (O) | Backhaul (B) | Duration Limit (L) | Time Window (TW) |
+ | :---------- | :----------: | :------------: | :----------: | :----------------: | :--------------: |
+ | CVRP | β | | | | |
+ | OVRP | β | β | | | |
+ | VRPB | β | | β | | |
+ | VRPL | β | | | β | |
+ | VRPTW | β | | | | β |
+ | OVRPTW | β | β | | | β |
+ | OVRPB | β | β | β | | |
+ | OVRPL | β | β | | β | |
+ | VRPBL | β | | β | β | |
+ | VRPBTW | β | | β | | β |
+ | VRPLTW | β | | | β | β |
+ | OVRPBL | β | β | β | β | |
+ | OVRPBTW | β | β | β | | β |
+ | OVRPLTW | β | β | | β | β |
+ | VRPBLTW | β | | β | β | β |
+ | OVRPBLTW | β | β | β | β | β |
You may also check out the following papers as reference:
- - `"Multi-Task Learning for Routing Problem with Cross-Problem Zero-Shot Generalization" (Liu et al, 2024) `_
- - `"MVMoE: Multi-Task Vehicle Routing Solver with Mixture-of-Experts" (Zhou et al, 2024) `_.
- - `"RouteFinder: Towards Foundation Models for Vehicle Routing Problems" (Berto et al, 2024) `_.
-
- Note:
+ - ["Multi-Task Learning for Routing Problem with Cross-Problem Zero-Shot Generalization" (Liu et al, 2024)](https://arxiv.org/abs/2402.16891)
+ - ["MVMoE: Multi-Task Vehicle Routing Solver with Mixture-of-Experts" (Zhou et al, 2024)](https://arxiv.org/abs/2405.01029)
+ - ["RouteFinder: Towards Foundation Models for Vehicle Routing Problems" (Berto et al, 2024)](https://arxiv.org/abs/2406.15007)
+
+ Tip:
Have a look at https://pyvrp.org/ for more information about VRP and its variants and their solutions. Kudos to their help and great job!
Args:
diff --git a/rl4co/envs/routing/pdp/env.py b/rl4co/envs/routing/pdp/env.py
index 6c46e4a3..dfa78904 100644
--- a/rl4co/envs/routing/pdp/env.py
+++ b/rl4co/envs/routing/pdp/env.py
@@ -235,7 +235,7 @@ def render(td: TensorDict, actions: torch.Tensor = None, ax=None):
class PDPRuinRepairEnv(ImprovementEnvBase):
- """Pickup and Delivery Problem (PDP) environment for performing neural rein-repair search.
+ """Pickup and Delivery Problem (PDP) environment for performing neural ruin-repair search.
The environment is made of num_loc + 1 locations (cities):
- 1 depot
- `num_loc` / 2 pickup locations
diff --git a/rl4co/envs/scheduling/ffsp/env.py b/rl4co/envs/scheduling/ffsp/env.py
index da8eff24..c74ef4c2 100644
--- a/rl4co/envs/scheduling/ffsp/env.py
+++ b/rl4co/envs/scheduling/ffsp/env.py
@@ -13,6 +13,7 @@
UnboundedDiscreteTensorSpec,
)
+from rl4co.data.dataset import FastTdDataset
from rl4co.envs.common.base import RL4COEnvBase
from .generator import FFSPGenerator
@@ -57,7 +58,7 @@ def __init__(
generator_params: dict = {},
**kwargs,
):
- super().__init__(check_solution=False, **kwargs)
+ super().__init__(check_solution=False, dataset_cls=FastTdDataset, **kwargs)
if generator is None:
generator = FFSPGenerator(**generator_params)
self.generator = generator
@@ -150,15 +151,12 @@ def pre_step(self, td: TensorDict) -> TensorDict:
batch_idx = torch.arange(*batch_size, dtype=torch.long, device=td.device)
sub_time_idx = td["sub_time_idx"]
# update machine index
- machine_idx = self.tables.get_machine_index(batch_idx, sub_time_idx)
+ td["machine_idx"] = self.tables.get_machine_index(batch_idx, sub_time_idx)
# update action mask and stage machine indx
td = self._update_step_state(td)
# perform some checks
- stage_machine_idx = td["stage_machine_idx"]
- stage_idx = td["stage_idx"]
- is_stage_one = stage_idx == 0
- assert torch.all(stage_machine_idx[is_stage_one] == machine_idx[is_stage_one])
- assert is_stage_one.all(), "call pre_step only at beginning of env"
+ assert (td["stage_idx"] == 0).all(), "call pre_step only at beginning of env"
+ assert torch.all(td["stage_machine_idx"] == td["machine_idx"])
# return updated td
return td
@@ -215,9 +213,6 @@ def _step(self, td: TensorDict) -> TensorDict:
time_idx = td["time_idx"]
machine_idx = td["machine_idx"]
- # create new td to avoid incplace ops and gradient problems resulting from this
- # td = td.clone()
-
# increment the operation counter of the selected job
td["job_location"][batch_idx, job_idx] += 1
# td["job_location"][:, :-1].clip_(0, self.num_stage)
@@ -311,21 +306,7 @@ def _reset(
dtype=torch.long,
device=device,
)
- if self.flatten_stages:
- assert (
- len(td["run_time"].shape) == 3
- ), "cost matrix has shape other than (bs, jobs, ma_total)"
- job_duration[..., : self.num_job, :] = td["run_time"]
- else:
- assert (
- len(td["run_time"].shape) == 4
- ), "cost matrix has shape other than (bs, jobs, ma, stages)"
- job_duration[..., : self.num_job, :] = (
- td["run_time"]
- .transpose(-2, -1)
- .contiguous()
- .view(*batch_size, self.num_job, self.num_machine_total)
- )
+ job_duration[..., : self.num_job, :] = td["run_time"]
job_duration[..., self.num_job, :] = 0
# Finish status information
@@ -429,16 +410,6 @@ def _make_spec(self, generator: FFSPGenerator):
def _get_reward(self, td, actions) -> TensorDict:
return td["reward"]
- def _get_cmap(self, color_cnt):
- from random import shuffle
-
- from matplotlib.colors import CSS4_COLORS, ListedColormap
-
- color_list = list(CSS4_COLORS.keys())
- shuffle(color_list)
- cmap = ListedColormap(color_list, N=color_cnt)
- return cmap
-
class IndexTables:
def __init__(self, env: FFSPEnv):
diff --git a/rl4co/envs/scheduling/ffsp/generator.py b/rl4co/envs/scheduling/ffsp/generator.py
index 8ec3ce71..6b33b8b1 100644
--- a/rl4co/envs/scheduling/ffsp/generator.py
+++ b/rl4co/envs/scheduling/ffsp/generator.py
@@ -1,17 +1,9 @@
-import os
-import zipfile
-from typing import Union, Callable
-
import torch
-import numpy as np
-from robust_downloader import download
-from torch.distributions import Uniform
from tensordict.tensordict import TensorDict
-from rl4co.data.utils import load_npz_to_tensordict
+from rl4co.envs.common.utils import Generator
from rl4co.utils.pylogger import get_pylogger
-from rl4co.envs.common.utils import get_sampler, Generator
log = get_pylogger(__name__)
@@ -34,6 +26,7 @@ class FFSPGenerator(Generator):
Note:
- [IMPORTANT] This version of ffsp requires the number of machines in each stage to be the same
"""
+
def __init__(
self,
num_stage: int = 2,
@@ -42,7 +35,7 @@ def __init__(
min_time: int = 2,
max_time: int = 10,
flatten_stages: bool = True,
- **unused_kwargs
+ **unused_kwargs,
):
self.num_stage = num_stage
self.num_machine = num_machine
@@ -61,16 +54,9 @@ def _generate(self, batch_size) -> TensorDict:
run_time = torch.randint(
low=self.min_time,
high=self.max_time,
- size=(*batch_size, self.num_job, self.num_machine, self.num_stage),
+ size=(*batch_size, self.num_job, self.num_machine_total),
)
- if self.flatten_stages:
- run_time = (
- run_time.transpose(-2, -1)
- .contiguous()
- .view(*batch_size, self.num_job, self.num_machine_total)
- )
-
return TensorDict(
{
"run_time": run_time,
diff --git a/rl4co/envs/scheduling/fjsp/env.py b/rl4co/envs/scheduling/fjsp/env.py
index dac1c8b6..dcf62608 100644
--- a/rl4co/envs/scheduling/fjsp/env.py
+++ b/rl4co/envs/scheduling/fjsp/env.py
@@ -79,14 +79,32 @@ def __init__(
else:
generator = FJSPGenerator(**generator_params)
self.generator = generator
- self.num_mas = generator.num_mas
- self.num_jobs = generator.num_jobs
- self.n_ops_max = generator.max_ops_per_job * self.num_jobs
+ self._num_mas = generator.num_mas
+ self._num_jobs = generator.num_jobs
+ self._n_ops_max = generator.max_ops_per_job * self.num_jobs
+
self.mask_no_ops = mask_no_ops
self.check_mask = check_mask
self.stepwise_reward = stepwise_reward
self._make_spec(self.generator)
+ @property
+ def num_mas(self):
+ return self._num_mas
+
+ @property
+ def num_jobs(self):
+ return self._num_jobs
+
+ @property
+ def n_ops_max(self):
+ return self._n_ops_max
+
+ def set_instance_params(self, td):
+ self._num_jobs = td["start_op_per_job"].size(1)
+ self._num_mas = td["proc_times"].size(1)
+ self._n_ops_max = td["proc_times"].size(2)
+
def _decode_graph_structure(self, td: TensorDict):
batch_size = td.batch_size
start_op_per_job = td["start_op_per_job"]
@@ -142,6 +160,8 @@ def _decode_graph_structure(self, td: TensorDict):
return td, n_ops_max
def _reset(self, td: TensorDict = None, batch_size=None) -> TensorDict:
+ self.set_instance_params(td)
+
td_reset = td.clone()
td_reset, n_ops_max = self._decode_graph_structure(td_reset)
@@ -333,10 +353,10 @@ def _make_step(self, td: TensorDict) -> TensorDict:
td["ops_sequence_order"] - gather_by_index(td["job_ops_adj"], selected_job, 1)
).clip(0)
# some checks
- assert torch.allclose(
- td["proc_times"].sum(1).gt(0).sum(1), # num ops with eligible machine
- (~(td["op_scheduled"] + td["pad_mask"])).sum(1), # num unscheduled ops
- )
+ # assert torch.allclose(
+ # td["proc_times"].sum(1).gt(0).sum(1), # num ops with eligible machine
+ # (~(td["op_scheduled"] + td["pad_mask"])).sum(1), # num unscheduled ops
+ # )
return td
@@ -483,7 +503,6 @@ def get_num_starts(self, td):
# NOTE in the paper they use N_s = 100
return 100
- @staticmethod
- def load_data(fpath, batch_size=[]):
+ def load_data(self, fpath, batch_size=[]):
g = FJSPFileGenerator(fpath)
return g(batch_size=batch_size)
diff --git a/rl4co/envs/scheduling/fjsp/generator.py b/rl4co/envs/scheduling/fjsp/generator.py
index 60246a50..8d2f427f 100644
--- a/rl4co/envs/scheduling/fjsp/generator.py
+++ b/rl4co/envs/scheduling/fjsp/generator.py
@@ -15,7 +15,6 @@
class FJSPGenerator(Generator):
-
"""Data generator for the Flexible Job-Shop Scheduling Problem (FJSP).
Args:
@@ -45,6 +44,7 @@ def __init__(
max_processing_time: int = 20,
min_eligible_ma_per_op: int = 1,
max_eligible_ma_per_op: int = None,
+ same_mean_per_op: bool = True,
**unused_kwargs,
):
self.num_jobs = num_jobs
@@ -58,7 +58,7 @@ def __init__(
# determines whether to use a fixed number of total operations or let it vary between instances
# NOTE: due to the way rl4co builds datasets, we need a fixed size here
self.n_ops_max = max_ops_per_job * num_jobs
-
+ self.same_mean_per_op = same_mean_per_op
# FFSP environment doen't have any other kwargs
if len(unused_kwargs) > 0:
log.error(f"Found {len(unused_kwargs)} unused kwargs: {unused_kwargs}")
@@ -86,12 +86,33 @@ def _simulate_processing_times(
ma_ops_edges = ma_ops_edges_unshuffled.gather(2, idx).transpose(1, 2)
# (bs, max_ops, machines)
- proc_times = torch.ones((bs, n_ops_max, self.num_mas))
- proc_times = torch.randint(
- self.min_processing_time,
- self.max_processing_time + 1,
- size=(bs, self.num_mas, n_ops_max),
- )
+ if self.same_mean_per_op:
+ proc_times = torch.ones((bs, self.num_mas, n_ops_max))
+ proc_time_means = torch.randint(
+ self.min_processing_time, self.max_processing_time, (bs, n_ops_max)
+ )
+ low_bounds = torch.maximum(
+ torch.full_like(proc_times, self.min_processing_time),
+ (proc_time_means * (1 - 0.2)).round().unsqueeze(1),
+ )
+ high_bounds = (
+ torch.minimum(
+ torch.full_like(proc_times, self.max_processing_time),
+ (proc_time_means * (1 + 0.2)).round().unsqueeze(1),
+ )
+ + 1
+ )
+ proc_times = (
+ torch.randint(2**63 - 1, size=proc_times.shape)
+ % (high_bounds - low_bounds)
+ + low_bounds
+ )
+ else:
+ proc_times = torch.randint(
+ self.min_processing_time,
+ self.max_processing_time + 1,
+ size=(bs, self.num_mas, n_ops_max),
+ )
# remove proc_times for which there is no corresponding ma-ops connection
proc_times = proc_times * ma_ops_edges
@@ -187,6 +208,8 @@ def __init__(self, file_path: str, n_ops_max: int = None, **unused_kwargs):
self.num_mas = num_machines
self.num_jobs = num_jobs
self.max_ops_per_job = max_ops_per_job
+ self.n_ops_max = max_ops_per_job * num_jobs
+
self.start_idx = 0
def _generate(self, batch_size: List[int]) -> TensorDict:
diff --git a/rl4co/models/__init__.py b/rl4co/models/__init__.py
index 16f27b18..339c3b01 100644
--- a/rl4co/models/__init__.py
+++ b/rl4co/models/__init__.py
@@ -43,6 +43,7 @@
from rl4co.models.zoo.n2s import N2S, N2SPolicy
from rl4co.models.zoo.nargnn import NARGNNPolicy
from rl4co.models.zoo.neuopt import NeuOpt, NeuOptPolicy
+from rl4co.models.zoo.polynet import PolyNet
from rl4co.models.zoo.pomo import POMO
from rl4co.models.zoo.ptrnet import PointerNetwork, PointerNetworkPolicy
from rl4co.models.zoo.symnco import SymNCO, SymNCOPolicy
diff --git a/rl4co/models/nn/attention.py b/rl4co/models/nn/attention.py
index 86217270..b65169f0 100644
--- a/rl4co/models/nn/attention.py
+++ b/rl4co/models/nn/attention.py
@@ -1,3 +1,4 @@
+import itertools
import math
import warnings
@@ -360,9 +361,18 @@ def _project_out(self, out, attn_mask):
num_nodes, num_available_nodes = attn_mask.size(-1), attn_mask.sum(-1)
# only do this at the "second" step, which is depot -> pomo -> first select
if (num_available_nodes >= num_nodes - 1).any():
- self.probs = F.softmax(self.dense_or_moe(out.view(-1, out.size(-1)).mean(dim=0, keepdim=True)), dim=-1)
+ self.probs = F.softmax(
+ self.dense_or_moe(
+ out.view(-1, out.size(-1)).mean(dim=0, keepdim=True)
+ ),
+ dim=-1,
+ )
selected = self.probs.multinomial(1).squeeze(0)
- out = self.project_out_moe(out) if selected.item() == 1 else self.project_out(out)
+ out = (
+ self.project_out_moe(out)
+ if selected.item() == 1
+ else self.project_out(out)
+ )
glimpse = out * self.probs.squeeze(0)[selected]
else:
glimpse = self.project_out_moe(out)
@@ -442,3 +452,86 @@ def forward(self, q, h=None, mask=None):
compatibility_s2n = torch.matmul(Q, K.transpose(2, 3))
return compatibility_s2n
+
+
+class PolyNetAttention(PointerAttention):
+ """Calculate logits given query, key and value and logit key.
+ This implements a modified version the pointer mechanism of Vinyals et al. (2015) (https://arxiv.org/abs/1506.03134)
+ as described in Hottung et al. (2024) (https://arxiv.org/abs/2402.14048) PolyNetAttention conditions the attention logits on
+ a set of k different binary vectors allowing to learn k different solution strategies.
+
+ Note:
+ With Flash Attention, masking is not supported
+
+ Performs the following:
+ 1. Apply cross attention to get the heads
+ 2. Project heads to get glimpse
+ 3. Apply PolyNet layers
+ 4. Compute attention score between glimpse and logit key
+
+ Args:
+ k: Number unique bit vectors used to compute attention score
+ embed_dim: total dimension of the model
+ poly_layer_dim: Dimension of the PolyNet layers
+ num_heads: number of heads
+ mask_inner: whether to mask inner attention
+ linear_bias: whether to use bias in linear projection
+ check_nan: whether to check for NaNs in logits
+ sdpa_fn: scaled dot product attention function (SDPA) implementation
+ """
+
+ def __init__(
+ self, k: int, embed_dim: int, poly_layer_dim: int, num_heads: int, **kwargs
+ ):
+ super(PolyNetAttention, self).__init__(embed_dim, num_heads, **kwargs)
+
+ self.k = k
+ self.binary_vector_dim = math.ceil(math.log2(k))
+ self.binary_vectors = torch.nn.Parameter(
+ torch.Tensor(
+ list(itertools.product([0, 1], repeat=self.binary_vector_dim))[:k]
+ ),
+ requires_grad=False,
+ )
+
+ self.poly_layer_1 = nn.Linear(embed_dim + self.binary_vector_dim, poly_layer_dim)
+ self.poly_layer_2 = nn.Linear(poly_layer_dim, embed_dim)
+
+ def forward(self, query, key, value, logit_key, attn_mask=None):
+ """Compute attention logits given query, key, value, logit key and attention mask.
+
+ Args:
+ query: query tensor of shape [B, ..., L, E]
+ key: key tensor of shape [B, ..., S, E]
+ value: value tensor of shape [B, ..., S, E]
+ logit_key: logit key tensor of shape [B, ..., S, E]
+ attn_mask: attention mask tensor of shape [B, ..., S]. Note that `True` means that the value _should_ take part in attention
+ as described in the [PyTorch Documentation](https://pytorch.org/docs/stable/generated/torch.nn.functional.scaled_dot_product_attention.html)
+ """
+ # Compute inner multi-head attention with no projections.
+ heads = self._inner_mha(query, key, value, attn_mask)
+ glimpse = self.project_out(heads)
+
+ num_solutions = glimpse.shape[1]
+ z = self.binary_vectors.repeat(math.ceil(num_solutions / self.k), 1)[
+ :num_solutions
+ ]
+ z = z[None].expand(glimpse.shape[0], num_solutions, self.binary_vector_dim)
+
+ # PolyNet layers
+ poly_out = self.poly_layer_1(torch.cat((glimpse, z), dim=2))
+ poly_out = F.relu(poly_out)
+ poly_out = self.poly_layer_2(poly_out)
+
+ glimpse += poly_out
+
+ # Batch matrix multiplication to compute logits (batch_size, num_steps, graph_size)
+ # bmm is slightly faster than einsum and matmul
+ logits = (torch.bmm(glimpse, logit_key.squeeze(-2).transpose(-2, -1))).squeeze(
+ -2
+ ) / math.sqrt(glimpse.size(-1))
+
+ if self.check_nan:
+ assert not torch.isnan(logits).any(), "Logits contain NaNs"
+
+ return logits
diff --git a/rl4co/models/nn/env_embeddings/init.py b/rl4co/models/nn/env_embeddings/init.py
index fa3b6fb6..06391cb2 100644
--- a/rl4co/models/nn/env_embeddings/init.py
+++ b/rl4co/models/nn/env_embeddings/init.py
@@ -407,6 +407,7 @@ def _op_features(self, td):
mean_durations = proc_times.sum(1) / (proc_times.gt(0).sum(1) + 1e-9)
feats = [
mean_durations / self.scaling_factor,
+ # td["lbs"] / self.scaling_factor,
td["is_ready"],
td["num_eligible"],
td["ops_job_map"],
@@ -430,20 +431,10 @@ def forward(self, td):
class FJSPInitEmbedding(JSSPInitEmbedding):
def __init__(self, embed_dim, linear_bias=False, scaling_factor: int = 100):
- super().__init__(embed_dim, linear_bias, scaling_factor, num_op_feats=5)
+ super().__init__(embed_dim, linear_bias, scaling_factor)
self.init_ma_embed = nn.Linear(1, self.embed_dim, bias=linear_bias)
self.edge_embed = nn.Linear(1, embed_dim, bias=linear_bias)
- def _op_features(self, td):
- feats = [
- td["lbs"] / self.scaling_factor,
- td["is_ready"],
- td["num_eligible"],
- td["op_scheduled"],
- td["ops_job_map"],
- ]
- return torch.stack(feats, dim=-1)
-
def forward(self, td: TensorDict):
ops_emb = self._init_ops_embed(td)
ma_emb = self._init_machine_embed(td)
@@ -471,19 +462,9 @@ def __init__(
linear_bias: bool = False,
scaling_factor: int = 1000,
):
- super().__init__(embed_dim, linear_bias, scaling_factor, num_op_feats=5)
+ super().__init__(embed_dim, linear_bias, scaling_factor)
self.init_ma_embed = nn.Linear(1, self.embed_dim, bias=linear_bias)
- def _op_features(self, td):
- feats = [
- td["lbs"] / self.scaling_factor,
- td["is_ready"],
- td["op_scheduled"],
- td["num_eligible"],
- td["ops_job_map"],
- ]
- return torch.stack(feats, dim=-1)
-
def _init_machine_embed(self, td: TensorDict):
busy_for = (td["busy_until"] - td["time"].unsqueeze(1)) / self.scaling_factor
ma_embeddings = self.init_ma_embed(busy_for.unsqueeze(2))
diff --git a/rl4co/models/rl/common/utils.py b/rl4co/models/rl/common/utils.py
index b23149f7..6c16976a 100644
--- a/rl4co/models/rl/common/utils.py
+++ b/rl4co/models/rl/common/utils.py
@@ -20,6 +20,8 @@ def __init__(self, scale: str = None):
def __call__(self, scores: torch.Tensor):
if self.scale is None:
return scores
+ elif isinstance(self.scale, int):
+ return scores / self.scale
# Score scaling
self.update(scores)
tensor_to_kwargs = dict(dtype=scores.dtype, device=scores.device)
diff --git a/rl4co/models/rl/ppo/__init__.py b/rl4co/models/rl/ppo/__init__.py
new file mode 100644
index 00000000..e69de29b
diff --git a/rl4co/models/rl/ppo/ppo.py b/rl4co/models/rl/ppo/ppo.py
index f8078036..7837b1e7 100644
--- a/rl4co/models/rl/ppo/ppo.py
+++ b/rl4co/models/rl/ppo/ppo.py
@@ -32,22 +32,22 @@ class PPO(RL4COLitModule):
The commonly implemented definition of policy entropy is the entropy of the policy distribution, given by:
- .. math:: H(\\pi(x_t)) = - \\sum_{a_t \\in A_t} \\pi(a_t|x_t) \\log \\pi(a_t|x_t)
+ $$H(\\pi(x_t)) = - \\sum_{a_t \\in A_t} \\pi(a_t|x_t) \\log \\pi(a_t|x_t)$$
- where :math:`x_t` represents the given state at step :math:`t`, :math:`A_t` is the set of all (admisible) actions
- at step :math:`t`, and :math:`a_t` is the action taken at step :math:`t`.
+ where $x_t$ represents the given state at step $t$, $A_t$ is the set of all (admisible) actions
+ at step $t$, and $a_t$ is the action taken at step $t$.
If we interpret autoregressive decoding steps as transition steps of an MDP, the entropy for the entire decoding
process can be defined as the sum of entropies for each decoding step:
- .. math:: H(\\pi) = \\sum_t H(\\pi(x_t))
+ $$H(\\pi) = \\sum_t H(\\pi(x_t))$$
However, if we consider autoregressive decoding steps as an algorithmic choice, the entropy for the entire decoding
process is defined as:
- .. math:: H(\\pi) = - \\sum_{a \\in A} \\pi(a|x) \\log \\pi(a|x)
+ $$H(\\pi) = - \\sum_{a \\in A} \\pi(a|x) \\log \\pi(a|x)$$
- where :math:`x` represents the given CO problem instance, and :math:`A` is the set of all feasible solutions.
+ where $x$ represents the given CO problem instance, and $A$ is the set of all feasible solutions.
Due to the intractability of computing the entropy of the policy distribution over all feasible solutions,
we approximate it by computing the entropy over solutions generated by the policy itself. This approximation serves
diff --git a/rl4co/models/rl/ppo/stepwise_ppo.py b/rl4co/models/rl/ppo/stepwise_ppo.py
index 98186ea1..49d087d0 100644
--- a/rl4co/models/rl/ppo/stepwise_ppo.py
+++ b/rl4co/models/rl/ppo/stepwise_ppo.py
@@ -1,13 +1,13 @@
import copy
-from typing import Any
+from typing import Any, Union
import torch
import torch.nn as nn
import torch.nn.functional as F
from torchrl.data.replay_buffers import (
- LazyTensorStorage,
+ LazyMemmapStorage,
ListStorage,
SamplerWithoutReplacement,
TensorDictReplayBuffer,
@@ -23,13 +23,17 @@
def make_replay_buffer(buffer_size, batch_size, device="cpu"):
if device == "cpu":
- storage = LazyTensorStorage(buffer_size, device="cpu")
+ storage = LazyMemmapStorage(buffer_size, device="cpu")
+ prefetch = 3
else:
storage = ListStorage(buffer_size)
+ prefetch = None
return TensorDictReplayBuffer(
storage=storage,
batch_size=batch_size,
sampler=SamplerWithoutReplacement(drop_last=True),
+ pin_memory=False,
+ prefetch=prefetch,
)
@@ -51,7 +55,7 @@ def __init__(
metrics: dict = {
"train": ["loss", "surrogate_loss", "value_loss", "entropy"],
},
- reward_scale: str = None,
+ reward_scale: Union[str, int] = None,
**kwargs,
):
super().__init__(env, policy, metrics=metrics, batch_size=batch_size, **kwargs)
@@ -143,13 +147,12 @@ def shared_step(
while not next_td["done"].all():
with torch.no_grad():
td = self.policy_old.act(next_td, self.env, phase="train")
-
- assert self.env._torchrl_mode, "Use torchrl mode in stepwise PPO"
- td = self.env.step(td)
- next_td = td.pop("next")
+ # get next state
+ next_td = self.env.step(td)["next"]
+ # get reward of action
reward = self.env.get_reward(next_td, None)
reward = self.scaler(reward)
-
+ # add reward to prior state
td.set("reward", reward)
# add tensordict with action, logprobs and reward information to buffer
self.rb.extend(td)
diff --git a/rl4co/models/zoo/__init__.py b/rl4co/models/zoo/__init__.py
index ba4345f3..7fbb41eb 100644
--- a/rl4co/models/zoo/__init__.py
+++ b/rl4co/models/zoo/__init__.py
@@ -24,6 +24,7 @@
from rl4co.models.zoo.n2s import N2S, N2SPolicy
from rl4co.models.zoo.nargnn import NARGNNPolicy
from rl4co.models.zoo.neuopt import NeuOpt, NeuOptPolicy
+from rl4co.models.zoo.polynet import PolyNet
from rl4co.models.zoo.pomo import POMO
from rl4co.models.zoo.ptrnet import PointerNetwork, PointerNetworkPolicy
from rl4co.models.zoo.symnco import SymNCO, SymNCOPolicy
diff --git a/rl4co/models/zoo/deepaco/antsystem.py b/rl4co/models/zoo/deepaco/antsystem.py
index d08df652..1965cd3d 100644
--- a/rl4co/models/zoo/deepaco/antsystem.py
+++ b/rl4co/models/zoo/deepaco/antsystem.py
@@ -273,7 +273,7 @@ def _update_pheromone(self, actions, reward):
self.pheromone += delta_pheromone
def _reward_map(self, x: Tensor):
- """Map reward :math:`f: \\mathbb{R} \\rightarrow \\mathbb{R}^+`"""
+ """Map reward $f: \\mathbb{R} \\rightarrow \\mathbb{R}^+$"""
M, _ = x.max(-1, keepdim=True)
m, _ = x.min(-1, keepdim=True)
v = ((x - m) / (M - m)) ** 2 * self.Q
diff --git a/rl4co/models/zoo/l2d/decoder.py b/rl4co/models/zoo/l2d/decoder.py
index b0ab3041..833e9c6e 100644
--- a/rl4co/models/zoo/l2d/decoder.py
+++ b/rl4co/models/zoo/l2d/decoder.py
@@ -178,7 +178,6 @@ def __init__(
actor_hidden_dim: int = 128,
actor_hidden_layers: int = 2,
num_encoder_layers: int = 3,
- num_heads: int = 8,
normalization: str = "batch",
het_emb: bool = False,
stepwise: bool = False,
diff --git a/rl4co/models/zoo/l2d/policy.py b/rl4co/models/zoo/l2d/policy.py
index b4b9b11c..0cfac356 100644
--- a/rl4co/models/zoo/l2d/policy.py
+++ b/rl4co/models/zoo/l2d/policy.py
@@ -35,6 +35,7 @@ def __init__(
env_name: str = "fjsp",
het_emb: bool = True,
scaling_factor: int = 1000,
+ normalization: str = "batch",
init_embedding: Optional[nn.Module] = None,
stepwise_encoding: bool = False,
tanh_clipping: float = 10,
@@ -77,6 +78,7 @@ def __init__(
het_emb=het_emb,
stepwise=stepwise_encoding,
scaling_factor=scaling_factor,
+ normalization=normalization,
)
# Pass to constructive policy
@@ -101,6 +103,7 @@ def __init__(
num_heads: int = 8,
num_encoder_layers: int = 4,
scaling_factor: int = 1000,
+ normalization: str = "batch",
env_name: str = "fjsp",
init_embedding: Optional[nn.Module] = None,
tanh_clipping: float = 10,
@@ -122,7 +125,7 @@ def __init__(
embed_dim=embed_dim,
num_heads=num_heads,
num_layers=num_encoder_layers,
- normalization="batch",
+ normalization=normalization,
feedforward_hidden=embed_dim * 2,
init_embedding=init_embedding,
)
diff --git a/rl4co/models/zoo/matnet/decoder.py b/rl4co/models/zoo/matnet/decoder.py
index ad8aef77..5a8d6e28 100644
--- a/rl4co/models/zoo/matnet/decoder.py
+++ b/rl4co/models/zoo/matnet/decoder.py
@@ -9,7 +9,8 @@
from rl4co.models.nn.env_embeddings.context import FFSPContext
from rl4co.models.zoo.am.decoder import AttentionModelDecoder
-from rl4co.utils.decoding import process_logits
+from rl4co.utils.decoding import decode_logprobs, process_logits
+from rl4co.utils.ops import gather_by_index
@dataclass
@@ -53,6 +54,8 @@ def __init__(
self,
embed_dim: int,
num_heads: int,
+ linear_bias: bool = False,
+ out_bias_pointer_attn: bool = True,
use_graph_context: bool = False,
**kwargs,
):
@@ -62,8 +65,10 @@ def __init__(
env_name="ffsp",
embed_dim=embed_dim,
num_heads=num_heads,
- use_graph_context=use_graph_context,
context_embedding=context_embedding,
+ out_bias_pointer_attn=out_bias_pointer_attn,
+ linear_bias=linear_bias,
+ use_graph_context=use_graph_context,
**kwargs,
)
@@ -116,6 +121,7 @@ def __init__(
embed_dim: int,
num_heads: int,
use_graph_context: bool = True,
+ tanh_clipping: float = 10,
**kwargs,
):
super().__init__(
@@ -125,7 +131,7 @@ def __init__(
**kwargs,
)
self.cached_embs: PrecomputedCache = None
- # self.encoded_wait_op = nn.Parameter(torch.rand((1, 1, embed_dim)))
+ self.tanh_clipping = tanh_clipping
def _precompute_cache(self, embeddings: Tuple[Tensor], **kwargs):
self.cached_embs = super()._precompute_cache(embeddings, **kwargs)
@@ -137,38 +143,15 @@ def forward(
num_starts: int = 1,
**decoding_kwargs,
) -> Tuple[Tensor, Tensor, TensorDict]:
- device = td.device
- batch_size = td.size(0)
- # TODO: we need to insert precompute cache inside the decoder
logits, mask = super().forward(td, self.cached_embs, num_starts)
logprobs = process_logits(
logits,
mask,
+ tanh_clipping=self.tanh_clipping,
**decoding_kwargs,
)
- all_job_probs = logprobs.exp()
-
- if "sampling" in decode_type:
- # to fix pytorch.multinomial bug on selecting 0 probability elements
- while True:
- job_selected = all_job_probs.multinomial(1).squeeze(dim=1)
- # shape: (batch)
- job_prob = all_job_probs.gather(1, job_selected[:, None]).squeeze(dim=1)
- # shape: (batch)
- assert (job_prob[td["done"].squeeze()] == 1).all()
-
- if (job_prob != 0).all():
- break
-
- elif "greedy" in decode_type:
- job_selected = all_job_probs.argmax(dim=1)
- # shape: (batch)
- job_prob = torch.zeros(
- size=(batch_size,), device=device
- ) # any number is okay
-
- else:
- raise ValueError(f"decode type {decode_type} not understood")
+ job_selected = decode_logprobs(logprobs, mask, decode_type)
+ job_prob = gather_by_index(logprobs, job_selected, dim=1)
return job_selected, job_prob
diff --git a/rl4co/models/zoo/matnet/encoder.py b/rl4co/models/zoo/matnet/encoder.py
index 3ad52309..0af88e23 100644
--- a/rl4co/models/zoo/matnet/encoder.py
+++ b/rl4co/models/zoo/matnet/encoder.py
@@ -6,7 +6,7 @@
from rl4co.models.nn.attention import MultiHeadCrossAttention
from rl4co.models.nn.env_embeddings import env_init_embedding
-from rl4co.models.nn.ops import Normalization
+from rl4co.models.nn.ops import TransformerFFN
class MixedScoresSDPA(nn.Module):
@@ -47,14 +47,15 @@ def forward(self, q, k, v, attn_mask=None, dmat=None, dropout_p=0.0):
# Calculate scaled dot product
attn_scores = torch.matmul(q, k.transpose(-2, -1)) / (k.size(-1) ** 0.5)
+ # [b, h, m, n, num_scores+1]
mix_attn_scores = torch.cat(
[
attn_scores.unsqueeze(-1),
dmat[:, None, ...].expand(b, self.num_heads, m, n, self.num_scores),
],
dim=-1,
- ) # [b, h, m, n, num_scores+1]
-
+ )
+ # [b, h, m, n]
attn_scores = (
(
torch.matmul(
@@ -68,7 +69,7 @@ def forward(self, q, k, v, attn_mask=None, dmat=None, dropout_p=0.0):
)
.transpose(1, 2)
.squeeze(-1)
- ) # [b, h, m, n]
+ )
# Apply the provided attention mask
if attn_mask is not None:
@@ -141,7 +142,7 @@ def forward(self, row_emb, col_emb, dmat, attn_mask=None):
return updated_row_emb, updated_col_emb
-class MatNetMHALayer(nn.Module):
+class MatNetLayer(nn.Module):
def __init__(
self,
embed_dim: int,
@@ -152,30 +153,8 @@ def __init__(
):
super().__init__()
self.MHA = MatNetMHA(embed_dim, num_heads, bias)
-
- self.F_a = nn.ModuleDict(
- {
- "norm1": Normalization(embed_dim, normalization),
- "ffn": nn.Sequential(
- nn.Linear(embed_dim, feedforward_hidden),
- nn.ReLU(),
- nn.Linear(feedforward_hidden, embed_dim),
- ),
- "norm2": Normalization(embed_dim, normalization),
- }
- )
-
- self.F_b = nn.ModuleDict(
- {
- "norm1": Normalization(embed_dim, normalization),
- "ffn": nn.Sequential(
- nn.Linear(embed_dim, feedforward_hidden),
- nn.ReLU(),
- nn.Linear(feedforward_hidden, embed_dim),
- ),
- "norm2": Normalization(embed_dim, normalization),
- }
- )
+ self.F_a = TransformerFFN(embed_dim, feedforward_hidden, normalization)
+ self.F_b = TransformerFFN(embed_dim, feedforward_hidden, normalization)
def forward(self, row_emb, col_emb, dmat, attn_mask=None):
"""
@@ -190,63 +169,18 @@ def forward(self, row_emb, col_emb, dmat, attn_mask=None):
"""
row_emb_out, col_emb_out = self.MHA(row_emb, col_emb, dmat, attn_mask)
-
- row_emb_out = self.F_a["norm1"](row_emb + row_emb_out)
- row_emb_out = self.F_a["norm2"](row_emb_out + self.F_a["ffn"](row_emb_out))
-
- col_emb_out = self.F_b["norm1"](col_emb + col_emb_out)
- col_emb_out = self.F_b["norm2"](col_emb_out + self.F_b["ffn"](col_emb_out))
+ row_emb_out = self.F_a(row_emb_out, row_emb)
+ col_emb_out = self.F_b(col_emb_out, col_emb)
return row_emb_out, col_emb_out
-class MatNetMHANetwork(nn.Module):
- def __init__(
- self,
- embed_dim: int = 128,
- num_heads: int = 8,
- num_layers: int = 3,
- normalization: str = "batch",
- feedforward_hidden: int = 512,
- bias: bool = False,
- ):
- super().__init__()
- self.layers = nn.ModuleList(
- [
- MatNetMHALayer(
- num_heads=num_heads,
- embed_dim=embed_dim,
- feedforward_hidden=feedforward_hidden,
- normalization=normalization,
- bias=bias,
- )
- for _ in range(num_layers)
- ]
- )
-
- def forward(self, row_emb, col_emb, dmat, attn_mask=None):
- """
- Args:
- row_emb (Tensor): [b, m, d]
- col_emb (Tensor): [b, n, d]
- dmat (Tensor): [b, m, n]
-
- Returns:
- Updated row_emb (Tensor): [b, m, d]
- Updated col_emb (Tensor): [b, n, d]
- """
-
- for layer in self.layers:
- row_emb, col_emb = layer(row_emb, col_emb, dmat, attn_mask)
- return row_emb, col_emb
-
-
class MatNetEncoder(nn.Module):
def __init__(
self,
embed_dim: int = 256,
num_heads: int = 16,
- num_layers: int = 5,
- normalization: str = "instance",
+ num_layers: int = 3,
+ normalization: str = "batch",
feedforward_hidden: int = 512,
init_embedding: nn.Module = None,
init_embedding_kwargs: dict = {},
@@ -261,15 +195,19 @@ def __init__(
)
self.init_embedding = init_embedding
- self.net = MatNetMHANetwork(
- embed_dim=embed_dim,
- num_heads=num_heads,
- num_layers=num_layers,
- normalization=normalization,
- feedforward_hidden=feedforward_hidden,
- bias=bias,
- )
self.mask_non_neighbors = mask_non_neighbors
+ self.layers = nn.ModuleList(
+ [
+ MatNetLayer(
+ embed_dim=embed_dim,
+ num_heads=num_heads,
+ bias=bias,
+ feedforward_hidden=feedforward_hidden,
+ normalization=normalization,
+ )
+ for _ in range(num_layers)
+ ]
+ )
def forward(self, td, attn_mask: torch.Tensor = None):
row_emb, col_emb, dmat = self.init_embedding(td)
@@ -278,7 +216,8 @@ def forward(self, td, attn_mask: torch.Tensor = None):
# attn_mask (keep 1s discard 0s) to only attend on neighborhood
attn_mask = dmat.ne(0)
- row_emb, col_emb = self.net(row_emb, col_emb, dmat, attn_mask)
+ for layer in self.layers:
+ row_emb, col_emb = layer(row_emb, col_emb, dmat, attn_mask)
embedding = (row_emb, col_emb)
init_embedding = None
diff --git a/rl4co/models/zoo/matnet/policy.py b/rl4co/models/zoo/matnet/policy.py
index 0c2af426..4e3ea980 100644
--- a/rl4co/models/zoo/matnet/policy.py
+++ b/rl4co/models/zoo/matnet/policy.py
@@ -96,15 +96,15 @@ class MultiStageFFSPPolicy(nn.Module):
def __init__(
self,
stage_cnt: int,
- embed_dim: int = 256,
+ embed_dim: int = 512,
num_heads: int = 16,
- num_encoder_layers: int = 3,
+ num_encoder_layers: int = 5,
use_graph_context: bool = False,
normalization: str = "instance",
feedforward_hidden: int = 512,
bias: bool = False,
train_decode_type: str = "sampling",
- val_decode_type: str = "sampling", # authors report better results for sampling
+ val_decode_type: str = "sampling",
test_decode_type: str = "sampling",
):
super().__init__()
@@ -136,8 +136,9 @@ def __init__(
self.test_decode_type = test_decode_type
def pre_forward(self, td: TensorDict, env: FFSPEnv, num_starts: int):
+ run_time_list = td["run_time"].chunk(env.num_stage, dim=-1)
for stage_idx in range(self.stage_cnt):
- td["cost_matrix"] = td["run_time"][:, :, :, stage_idx]
+ td["cost_matrix"] = run_time_list[stage_idx]
encoder = self.encoders[stage_idx]
embeddings, _ = encoder(td)
decoder = self.decoders[stage_idx]
@@ -171,36 +172,36 @@ def forward(
# NOTE: this must come after pre_forward due to batchify op
batch_size = td.size(0)
- prob_list = torch.zeros(size=(batch_size, 0), device=device)
+ logp_list = torch.zeros(size=(batch_size, 0), device=device)
action_list = []
while not td["done"].all():
action_stack = torch.empty(
size=(batch_size, self.stage_cnt), dtype=torch.long, device=device
)
- prob_stack = torch.empty(size=(batch_size, self.stage_cnt), device=device)
+ logp_stack = torch.empty(size=(batch_size, self.stage_cnt), device=device)
for stage_idx in range(self.stage_cnt):
decoder = self.decoders[stage_idx]
- action, prob = decoder(td, decode_type, num_starts, **decoder_kwargs)
+ action, logp = decoder(td, decode_type, num_starts, **decoder_kwargs)
action_stack[:, stage_idx] = action
- prob_stack[:, stage_idx] = prob
+ logp_stack[:, stage_idx] = logp
gathering_index = td["stage_idx"][:, None]
# shape: (batch, 1)
action = action_stack.gather(dim=1, index=gathering_index).squeeze(dim=1)
- prob = prob_stack.gather(dim=1, index=gathering_index).squeeze(dim=1)
+ logp = logp_stack.gather(dim=1, index=gathering_index).squeeze(dim=1)
# shape: (batch)
action_list.append(action)
# transition
td.set("action", action)
td = env.step(td)["next"]
- prob_list = torch.cat((prob_list, prob[:, None]), dim=1)
+ logp_list = torch.cat((logp_list, logp[:, None]), dim=1)
out = {
"reward": td["reward"],
- "log_likelihood": prob_list.log().sum(1),
+ "log_likelihood": logp_list.sum(1),
}
if return_actions:
diff --git a/rl4co/models/zoo/polynet/__init__.py b/rl4co/models/zoo/polynet/__init__.py
new file mode 100644
index 00000000..1a908241
--- /dev/null
+++ b/rl4co/models/zoo/polynet/__init__.py
@@ -0,0 +1 @@
+from .model import PolyNet
diff --git a/rl4co/models/zoo/polynet/decoder.py b/rl4co/models/zoo/polynet/decoder.py
new file mode 100644
index 00000000..7a28fa1a
--- /dev/null
+++ b/rl4co/models/zoo/polynet/decoder.py
@@ -0,0 +1,145 @@
+from dataclasses import dataclass
+from typing import Tuple, Union
+
+import torch.nn as nn
+
+from torch import Tensor
+
+from rl4co.envs import RL4COEnvBase
+from rl4co.models.nn.attention import PolyNetAttention
+from rl4co.models.nn.env_embeddings import env_context_embedding, env_dynamic_embedding
+from rl4co.models.nn.env_embeddings.dynamic import StaticEmbedding
+from rl4co.models.zoo.am.decoder import AttentionModelDecoder
+from rl4co.utils.pylogger import get_pylogger
+
+log = get_pylogger(__name__)
+
+
+@dataclass
+class PrecomputedCache:
+ node_embeddings: Tensor
+ graph_context: Union[Tensor, float]
+ glimpse_key: Tensor
+ glimpse_val: Tensor
+ logit_key: Tensor
+
+
+class PolyNetDecoder(AttentionModelDecoder):
+ """
+ PolyNet decoder for constructing diverse solutions for combinatorial optimization problems.
+ Given the environment state and the embeddings, compute the logits and sample actions autoregressively until
+ all the environments in the batch have reached a terminal state.
+ We additionally include support for multi-starts as it is more efficient to do so in the decoder as we can
+ natively perform the attention computation.
+
+ Args:
+ k: Number of strategies to learn ("K" in the PolyNet paper)
+ encoder_type: Type of encoder that should be used. "AM" or "MatNet" are supported
+ embed_dim: Embedding dimension
+ poly_layer_dim: Dimension of the PolyNet layers
+ num_heads: Number of attention heads
+ env_name: Name of the environment used to initialize embeddings
+ context_embedding: Context embedding module
+ dynamic_embedding: Dynamic embedding module
+ mask_inner: Whether to mask the inner loop
+ out_bias_pointer_attn: Whether to use a bias in the pointer attention
+ linear_bias: Whether to use a bias in the linear layer
+ use_graph_context: Whether to use the graph context
+ check_nan: Whether to check for nan values during decoding
+ sdpa_fn: scaled_dot_product_attention function
+ """
+
+ def __init__(
+ self,
+ k: int,
+ encoder_type: str,
+ embed_dim: int = 128,
+ poly_layer_dim: int = 256,
+ num_heads: int = 8,
+ env_name: Union[str, RL4COEnvBase] = "tsp",
+ context_embedding: nn.Module = None,
+ dynamic_embedding: nn.Module = None,
+ mask_inner: bool = True,
+ out_bias_pointer_attn: bool = False,
+ linear_bias: bool = False,
+ use_graph_context: bool = True,
+ check_nan: bool = True,
+ sdpa_fn: callable = None,
+ **unused_kwargs,
+ ):
+ super().__init__()
+
+ if isinstance(env_name, RL4COEnvBase):
+ env_name = env_name.name
+ self.env_name = env_name
+ self.embed_dim = embed_dim
+ self.num_heads = num_heads
+ self.encoder_type = encoder_type
+
+ assert embed_dim % num_heads == 0
+
+ self.context_embedding = (
+ env_context_embedding(self.env_name, {"embed_dim": embed_dim})
+ if context_embedding is None
+ else context_embedding
+ )
+ self.dynamic_embedding = (
+ env_dynamic_embedding(self.env_name, {"embed_dim": embed_dim})
+ if dynamic_embedding is None
+ else dynamic_embedding
+ )
+ self.is_dynamic_embedding = (
+ False if isinstance(self.dynamic_embedding, StaticEmbedding) else True
+ )
+
+ # MHA with Pointer mechanism (https://arxiv.org/abs/1506.03134)
+ self.pointer = PolyNetAttention(
+ k,
+ embed_dim,
+ poly_layer_dim,
+ num_heads,
+ mask_inner=mask_inner,
+ out_bias=out_bias_pointer_attn,
+ check_nan=check_nan,
+ sdpa_fn=sdpa_fn,
+ )
+
+ # For each node we compute (glimpse key, glimpse value, logit key) so 3 * embed_dim
+ self.project_node_embeddings = nn.Linear(
+ embed_dim, 3 * embed_dim, bias=linear_bias
+ )
+ self.project_fixed_context = nn.Linear(embed_dim, embed_dim, bias=linear_bias)
+ self.use_graph_context = use_graph_context
+
+ def _precompute_cache_matnet(
+ self, embeddings: Tuple[Tensor, Tensor], *args, **kwargs
+ ):
+ col_emb, row_emb = embeddings
+ (
+ glimpse_key_fixed,
+ glimpse_val_fixed,
+ logit_key,
+ ) = self.project_node_embeddings(
+ col_emb
+ ).chunk(3, dim=-1)
+
+ # Optionally disable the graph context from the initial embedding as done in POMO
+ if self.use_graph_context:
+ graph_context = self.project_fixed_context(col_emb.mean(1))
+ else:
+ graph_context = 0
+
+ # Organize in a dataclass for easy access
+ return PrecomputedCache(
+ node_embeddings=row_emb,
+ graph_context=graph_context,
+ glimpse_key=glimpse_key_fixed,
+ glimpse_val=glimpse_val_fixed,
+ logit_key=logit_key,
+ )
+
+ def _precompute_cache(self, embeddings: Tuple[Tensor, Tensor], *args, **kwargs):
+ if self.encoder_type == "AM":
+ return super()._precompute_cache(embeddings, *args, **kwargs)
+ elif self.encoder_type == "MatNet":
+ return self._precompute_cache_matnet(embeddings, *args, **kwargs)
diff --git a/rl4co/models/zoo/polynet/model.py b/rl4co/models/zoo/polynet/model.py
new file mode 100644
index 00000000..a1271da8
--- /dev/null
+++ b/rl4co/models/zoo/polynet/model.py
@@ -0,0 +1,241 @@
+import logging
+
+from typing import Any, Optional, Union
+
+import torch
+
+from tensordict import TensorDict
+
+from rl4co.data.transforms import StateAugmentation
+from rl4co.envs.common.base import RL4COEnvBase
+from rl4co.models.rl.reinforce.reinforce import REINFORCE
+from rl4co.models.zoo.polynet.policy import PolyNetPolicy
+from rl4co.utils.ops import gather_by_index, unbatchify
+from rl4co.utils.pylogger import get_pylogger
+
+log = get_pylogger(__name__)
+
+
+class PolyNet(REINFORCE):
+ """PolyNet
+ Based on Hottung et al. (2024) https://arxiv.org/abs/2402.14048.
+
+ Note:
+ PolyNet allows to learn diverse solution stratgies with a single model. This is achieved
+ through a modified decoder and the Poppy loss (Grinsztajn et al. (2021)). PolyNet can be used with the attention model encoder or the MatNet encoder by
+ setting encoder_type to "AM" or "MatNet", respectively.
+
+ Args:
+ env: TorchRL Environment
+ policy: Policy to use for the algorithm
+ k: Number of strategies to learn ("K" in the paper)
+ val_num_solutions: Number of solutions that are generated per instance during validation
+ encoder_type: Type of encoder that should be used. "AM" or "MatNet" are supported
+ policy_kwargs: Keyword arguments for policy
+ baseline: Baseline to use for the algorithm. Note that PolyNet only supports shared baseline,
+ so we will throw an error if anything else is passed.
+ num_augment: Number of augmentations (used only for validation and test)
+ augment_fn: Function to use for augmentation, defaulting to dihedral8
+ first_aug_identity: Whether to include the identity augmentation in the first position
+ feats: List of features to augment
+ **kwargs: Keyword arguments passed to the superclass
+ """
+
+ def __init__(
+ self,
+ env: RL4COEnvBase,
+ policy: PolyNetPolicy = None,
+ k: int = 128,
+ val_num_solutions: int = 800,
+ encoder_type="AM",
+ base_model_checkpoint_path: str = None,
+ policy_kwargs={},
+ baseline: str = "shared",
+ num_augment: int = 8,
+ augment_fn: Union[str, callable] = "dihedral8",
+ first_aug_identity: bool = True,
+ feats: list = None,
+ **kwargs,
+ ):
+ self.save_hyperparameters(logger=False)
+
+ self.k = k
+ self.val_num_solutions = val_num_solutions
+
+ assert encoder_type in [
+ "AM",
+ "MatNet",
+ ], "Supported encoder types are 'AM' and 'MatNet'"
+
+ assert baseline == "shared", "PolyNet only supports shared baseline"
+
+ if (
+ policy_kwargs.get("val_decode_type") == "greedy"
+ or policy_kwargs.get("test_decode_type") == "greedy"
+ ):
+ assert (
+ val_num_solutions <= k
+ ), "If greedy decoding is used val_num_solutions must be <= k"
+
+ if encoder_type == "MatNet":
+ assert (
+ num_augment == 1
+ ), "MatNet does not use symmetric or dihedral augmentation"
+
+ if policy is None:
+ policy = PolyNetPolicy(
+ env_name=env.name, k=k, encoder_type=encoder_type, **policy_kwargs
+ )
+
+ if base_model_checkpoint_path is not None:
+ logging.info(
+ f"Trying to load weights from baseline model {base_model_checkpoint_path}"
+ )
+ checkpoint = torch.load(base_model_checkpoint_path)
+ state_dict = checkpoint["state_dict"]
+ state_dict = {k.replace("policy.", "", 1): v for k, v in state_dict.items()}
+ policy.load_state_dict(state_dict, strict=False)
+
+ train_batch_size = kwargs["batch_size"] if "batch_size" in kwargs else 64
+ kwargs_with_defaults = {
+ "val_batch_size": train_batch_size,
+ "test_batch_size": train_batch_size,
+ }
+ kwargs_with_defaults.update(kwargs)
+
+ # Initialize with the shared baseline
+ super(PolyNet, self).__init__(env, policy, baseline, **kwargs_with_defaults)
+
+ self.num_augment = num_augment
+ if self.num_augment > 1:
+ self.augment = StateAugmentation(
+ num_augment=self.num_augment,
+ augment_fn=augment_fn,
+ first_aug_identity=first_aug_identity,
+ feats=feats,
+ )
+ else:
+ self.augment = None
+
+ # Add `_multistart` to decode type for train, val and test in policy
+ # for phase in ["train", "val", "test"]:
+ # self.set_decode_type_multistart(phase)
+
+ def shared_step(
+ self, batch: Any, batch_idx: int, phase: str, dataloader_idx: int = None
+ ):
+ td = self.env.reset(batch)
+ n_aug = self.num_augment
+
+ # During training, we do not augment the data
+ if phase == "train":
+ n_aug = 0
+ elif n_aug > 1:
+ td = self.augment(td)
+
+ if phase == "train":
+ n_start = self.k
+ else:
+ n_start = self.val_num_solutions
+
+ # Evaluate policy
+ out = self.policy(
+ td,
+ self.env,
+ phase=phase,
+ num_starts=n_start,
+ multisample=True,
+ return_actions=True,
+ )
+
+ # Unbatchify reward to [batch_size, num_augment, num_starts].
+ reward = unbatchify(out["reward"], (n_aug, n_start))
+
+ # Training phase
+ if phase == "train":
+ assert n_start > 1, "num_starts must be > 1 during training"
+ log_likelihood = unbatchify(out["log_likelihood"], (n_aug, n_start))
+ self.calculate_loss(td, batch, out, reward, log_likelihood)
+ max_reward, max_idxs = reward.max(dim=-1)
+ out.update({"max_reward": max_reward})
+ # Get multi-start (=POMO) rewards and best actions only during validation and test
+ else:
+ if n_start > 1:
+ # max multi-start reward
+ max_reward, max_idxs = reward.max(dim=-1)
+ out.update({"max_reward": max_reward})
+
+ if out.get("actions", None) is not None:
+ # Reshape batch to [batch_size, num_augment, num_starts, ...]
+ actions = unbatchify(out["actions"], (n_aug, n_start))
+ out.update(
+ {
+ "best_multistart_actions": gather_by_index(
+ actions, max_idxs.unsqueeze(2), dim=2
+ )
+ }
+ )
+ out["actions"] = actions
+
+ # Get augmentation score only during inference
+ if n_aug > 1:
+ # If multistart is enabled, we use the best multistart rewards
+ reward_ = max_reward if n_start > 1 else reward
+ max_aug_reward, max_idxs = reward_.max(dim=1)
+ out.update({"max_aug_reward": max_aug_reward})
+
+ if out.get("actions", None) is not None:
+ actions_ = (
+ out["best_multistart_actions"] if n_start > 1 else out["actions"]
+ )
+ out.update({"best_aug_actions": gather_by_index(actions_, max_idxs)})
+
+ metrics = self.log_metrics(out, phase, dataloader_idx=dataloader_idx)
+ return {"loss": out.get("loss", None), **metrics}
+
+ def calculate_loss(
+ self,
+ td: TensorDict,
+ batch: TensorDict,
+ policy_out: dict,
+ reward: Optional[torch.Tensor] = None,
+ log_likelihood: Optional[torch.Tensor] = None,
+ ):
+ """Calculate loss following Poppy (https://arxiv.org/abs/2210.03475).
+
+ Args:
+ td: TensorDict containing the current state of the environment
+ batch: Batch of data. This is used to get the extra loss terms, e.g., REINFORCE baseline
+ policy_out: Output of the policy network
+ reward: Reward tensor. If None, it is taken from `policy_out`
+ log_likelihood: Log-likelihood tensor. If None, it is taken from `policy_out`
+ """
+ # Extra: this is used for additional loss terms, e.g., REINFORCE baseline
+ extra = batch.get("extra", None)
+ reward = reward if reward is not None else policy_out["reward"]
+ log_likelihood = (
+ log_likelihood if log_likelihood is not None else policy_out["log_likelihood"]
+ )
+
+ # REINFORCE baseline
+ bl_val, bl_loss = (
+ self.baseline.eval(td, reward, self.env) if extra is None else (extra, 0)
+ )
+
+ # Log-likelihood mask. Mask everything but the best rollout per instance
+ best_idx = (-reward).argsort(1).argsort(1)
+ mask = best_idx < 1
+
+ # Main loss function
+ advantage = reward - bl_val # advantage = reward - baseline
+ reinforce_loss = -(advantage * log_likelihood * mask).mean()
+ loss = reinforce_loss + bl_loss
+ policy_out.update(
+ {
+ "loss": loss,
+ "reinforce_loss": reinforce_loss,
+ "bl_loss": bl_loss,
+ "bl_val": bl_val,
+ }
+ )
+ return policy_out
diff --git a/rl4co/models/zoo/polynet/policy.py b/rl4co/models/zoo/polynet/policy.py
new file mode 100644
index 00000000..d2128a6f
--- /dev/null
+++ b/rl4co/models/zoo/polynet/policy.py
@@ -0,0 +1,101 @@
+from typing import Union
+
+import torch.nn as nn
+
+from rl4co.envs import RL4COEnvBase
+from rl4co.models.common.constructive.autoregressive.policy import AutoregressivePolicy
+from rl4co.models.zoo.am.encoder import AttentionModelEncoder
+from rl4co.models.zoo.matnet.encoder import MatNetEncoder
+from rl4co.models.zoo.polynet.decoder import PolyNetDecoder
+
+
+class PolyNetPolicy(AutoregressivePolicy):
+ """
+ # TODO
+ Polynet policy based on Hottung et al. (2024) https://arxiv.org/abs/2402.14048.
+ The model uses either the AttentionModel encoder or the MatNet encoder in combination with
+ a custom PolyNet decoder.
+
+ Note: The default arguments for the AttentionModel encoder follow the POMO paper. The default decoding type
+ during validation and testing is 'sampling'.
+
+ Args:
+ k: Number of strategies to learn ("K" in the paper)
+ encoder_type: Type of encoder that should be used. "AM" or "MatNet" are supported.
+ embed_dim: Dimension of the node embeddings
+ num_encoder_layers: Number of layers in the encoder
+ num_heads: Number of heads in the attention layers
+ normalization: Normalization type in the attention layers
+ feedforward_hidden: Dimension of the hidden layer in the feedforward network
+ env_name: Name of the environment used to initialize embeddings
+ temperature: Temperature for the softmax
+ tanh_clipping: Tanh clipping value (see Bello et al., 2016)
+ mask_logits: Whether to mask the logits during decoding
+ train_decode_type: Type of decoding to use during training
+ val_decode_type: Type of decoding to use during validation
+ test_decode_type: Type of decoding to use during testing
+ **kwargs: keyword arguments passed to the encoder and decoder modules
+ """
+
+ def __init__(
+ self,
+ k: int,
+ encoder: nn.Module = None,
+ encoder_type: str = "AM",
+ embed_dim: int = 128,
+ num_encoder_layers: int = 6,
+ num_heads: int = 8,
+ normalization: str = "instance",
+ feedforward_hidden: int = 512,
+ env_name: Union[str, RL4COEnvBase] = "tsp",
+ temperature: float = 1.0,
+ tanh_clipping: float = 10.0,
+ mask_logits: bool = True,
+ train_decode_type: str = "sampling",
+ val_decode_type: str = "sampling",
+ test_decode_type: str = "sampling",
+ **kwargs,
+ ):
+ if encoder is None:
+ if encoder_type == "AM":
+ encoder = AttentionModelEncoder(
+ embed_dim=embed_dim,
+ num_heads=num_heads,
+ num_layers=num_encoder_layers,
+ env_name=env_name,
+ normalization=normalization,
+ feedforward_hidden=feedforward_hidden,
+ **kwargs,
+ )
+ elif encoder_type == "MatNet":
+ kwargs_with_defaults = {"init_embedding_kwargs": {"mode": "RandomOneHot"}}
+ kwargs_with_defaults.update(kwargs)
+ encoder = MatNetEncoder(
+ embed_dim=embed_dim,
+ num_heads=num_heads,
+ num_layers=num_encoder_layers,
+ normalization=normalization,
+ **kwargs_with_defaults,
+ )
+
+ decoder = PolyNetDecoder(
+ k=k,
+ encoder_type=encoder_type,
+ embed_dim=embed_dim,
+ num_heads=num_heads,
+ env_name=env_name,
+ **kwargs,
+ )
+
+ super(PolyNetPolicy, self).__init__(
+ encoder=encoder,
+ decoder=decoder,
+ env_name=env_name,
+ temperature=temperature,
+ tanh_clipping=tanh_clipping,
+ mask_logits=mask_logits,
+ train_decode_type=train_decode_type,
+ val_decode_type=val_decode_type,
+ test_decode_type=test_decode_type,
+ **kwargs,
+ )
diff --git a/rl4co/tasks/eval.py b/rl4co/tasks/eval.py
index 10dda554..7580d374 100644
--- a/rl4co/tasks/eval.py
+++ b/rl4co/tasks/eval.py
@@ -51,17 +51,17 @@ def __call__(self, policy, dataloader, **kwargs):
actions_list.append(actions)
rewards = torch.cat(rewards_list)
-
+
# Padding: pad actions to the same length with zeros
- max_length = max(action.size(-1) for action in actions)
+ max_length = max(action.size(-1) for action in actions_list)
actions = torch.cat(
[
torch.nn.functional.pad(action, (0, max_length - action.size(-1)))
- for action in actions
+ for action in actions_list
],
0,
)
-
+
inference_time = time.time() - start
tqdm.write(f"Mean reward for {self.name}: {rewards.mean():.4f}")
diff --git a/rl4co/utils/decoding.py b/rl4co/utils/decoding.py
index 35193555..f29e8214 100644
--- a/rl4co/utils/decoding.py
+++ b/rl4co/utils/decoding.py
@@ -263,7 +263,7 @@ def pre_decoder_hook(
self, td: TensorDict, env: RL4COEnvBase, action: torch.Tensor = None
):
"""Pre decoding hook. This method is called before the main decoding operation."""
-
+
# Multi-start decoding. If num_starts is None, we use the number of actions in the action mask
if self.multistart or self.multisample:
if self.num_starts is None:
diff --git a/rl4co/utils/meta_trainer.py b/rl4co/utils/meta_trainer.py
new file mode 100644
index 00000000..ccd64352
--- /dev/null
+++ b/rl4co/utils/meta_trainer.py
@@ -0,0 +1,170 @@
+import lightning.pytorch as pl
+import torch
+import math
+import copy
+from torch.optim import Adam
+
+from lightning import Callback
+from rl4co import utils
+import random
+log = utils.get_pylogger(__name__)
+
+
+class ReptileCallback(Callback):
+
+ """ Meta training framework for addressing the generalization issue (implement the Reptile algorithm only)
+ Based on Manchanda et al. 2022 (https://arxiv.org/abs/2206.00787) and Zhou et al. 2023 (https://arxiv.org/abs/2305.19587)
+
+ Args:
+ - num_tasks: the number of tasks in a mini-batch, i.e. `B` in the original paper
+ - alpha: initial weight of the task model for the outer-loop optimization of reptile
+ - alpha_decay: weight decay of the task model for the outer-loop optimization of reptile
+ - min_size: minimum problem size of the task (only supported in cross-size generalization)
+ - max_size: maximum problem size of the task (only supported in cross-size generalization)
+ - sch_bar: for the task scheduler of size setting, where lr_decay_epoch = sch_bar * epochs, i.e. after this epoch, learning rate will decay with a weight 0.1
+ - data_type: type of the tasks, chosen from ["size", "distribution", "size_distribution"]
+ - print_log: whether to print the specific task sampled in each inner-loop optimization
+ """
+ def __init__(self,
+ num_tasks: int,
+ alpha: float,
+ alpha_decay: float,
+ min_size: int,
+ max_size: int,
+ sch_bar: float = 0.9,
+ data_type: str = "size",
+ print_log: bool =True):
+
+ super().__init__()
+
+ self.num_tasks = num_tasks
+ self.alpha = alpha
+ self.alpha_decay = alpha_decay
+ self.sch_bar = sch_bar
+ self.print_log = print_log
+ self.data_type = data_type
+ self.task_set = self._generate_task_set(data_type, min_size, max_size)
+
+ def on_fit_start(self, trainer: pl.Trainer, pl_module: pl.LightningModule) -> None:
+
+ # Sample a batch of tasks
+ self._sample_task()
+
+ # Pre-set the distribution
+ if self.data_type == "size_distribution":
+ pl_module.env.generator.loc_distribution = "gaussian_mixture"
+ self.selected_tasks[0] = (pl_module.env.generator.num_loc, 0, 0)
+ elif self.data_type == "size":
+ pl_module.env.generator.loc_distribution = "uniform"
+ self.selected_tasks[0] = (pl_module.env.generator.num_loc, )
+ elif self.data_type == "distribution":
+ pl_module.env.generator.loc_distribution = "gaussian_mixture"
+ self.selected_tasks[0] = (0, 0)
+ self.task_params = self.selected_tasks[0]
+
+ def on_train_epoch_start(self, trainer: pl.Trainer, pl_module: pl.LightningModule) -> None:
+
+ # Alpha scheduler (decay for the update of meta model)
+ self._alpha_scheduler()
+
+ # Reinitialize the task model with the parameters of the meta model
+ if trainer.current_epoch % self.num_tasks == 0: # Save the meta model
+ self.meta_model_state_dict = copy.deepcopy(pl_module.state_dict())
+ self.task_models = []
+ # Print sampled tasks
+ if self.print_log:
+ print('\n>> Meta epoch: {} (Exact epoch: {}), Training task: {}'.format(trainer.current_epoch//self.num_tasks, trainer.current_epoch, self.selected_tasks))
+ else:
+ pl_module.load_state_dict(self.meta_model_state_dict)
+
+ # Reinitialize the optimizer every epoch
+ lr_decay = 0.1 if trainer.current_epoch+1 == int(self.sch_bar * trainer.max_epochs) else 1
+ old_lr = trainer.optimizers[0].param_groups[0]['lr']
+ new_optimizer = Adam(pl_module.parameters(), lr=old_lr * lr_decay)
+ trainer.optimizers = [new_optimizer]
+
+ # Print
+ if self.print_log:
+ if hasattr(pl_module.env.generator, 'capacity'):
+ print('>> Training task: {}, capacity: {}'.format(self.task_params, pl_module.env.generator.capacity))
+ else:
+ print('>> Training task: {}'.format(self.task_params))
+
+ def on_train_epoch_end(self, trainer: pl.Trainer, pl_module: pl.LightningModule):
+
+ # Save the task model
+ self.task_models.append(copy.deepcopy(pl_module.state_dict()))
+ if (trainer.current_epoch+1) % self.num_tasks == 0:
+ # Outer-loop optimization (update the meta model with the parameters of the task model)
+ with torch.no_grad():
+ state_dict = {params_key: (self.meta_model_state_dict[params_key] +
+ self.alpha * torch.mean(torch.stack([fast_weight[params_key] - self.meta_model_state_dict[params_key]
+ for fast_weight in self.task_models], dim=0).float(), dim=0))
+ for params_key in self.meta_model_state_dict}
+ pl_module.load_state_dict(state_dict)
+
+ # Get ready for the next meta-training iteration
+ if (trainer.current_epoch + 1) % self.num_tasks == 0:
+ # Sample a batch of tasks
+ self._sample_task()
+
+ # Load new training task (Update the environment) for the next meta-training iteration
+ self._load_task(pl_module, task_idx = (trainer.current_epoch+1) % self.num_tasks)
+
+ def _sample_task(self):
+
+ # Sample a batch of tasks
+ self.selected_tasks = []
+ for b in range(self.num_tasks):
+ task_params = random.sample(self.task_set, 1)[0]
+ self.selected_tasks.append(task_params)
+
+ def _load_task(self, pl_module: pl.LightningModule, task_idx=0):
+
+ # Load new training task (Update the environment)
+ self.task_params = self.selected_tasks[task_idx]
+
+ if self.data_type == "size_distribution":
+ assert len(self.task_params) == 3
+ pl_module.env.generator.num_loc = self.task_params[0]
+ pl_module.env.generator.num_modes = self.task_params[1]
+ pl_module.env.generator.cdist = self.task_params[2]
+ elif self.data_type == "distribution": # fixed size
+ assert len(self.task_params) == 2
+ pl_module.env.generator.num_modes = self.task_params[0]
+ pl_module.env.generator.cdist = self.task_params[1]
+ elif self.data_type == "size": # fixed distribution
+ assert len(self.task_params) == 1
+ pl_module.env.generator.num_loc = self.task_params[0]
+
+ if hasattr(pl_module.env.generator, 'capacity') and self.data_type in ["size_distribution", "size"]:
+ task_capacity = math.ceil(30 + self.task_params[0] / 5) if self.task_params[0] >= 20 else 20
+ pl_module.env.generator.capacity = task_capacity
+
+ def _alpha_scheduler(self):
+ self.alpha = max(self.alpha * self.alpha_decay, 0.0001)
+
+ def _generate_task_set(self, data_type, min_size, max_size):
+ """
+ Following the setting in Zhou et al. 2023 (https://arxiv.org/abs/2305.19587)
+ Current setting:
+ size: (n,) \in [20, 150]
+ distribution: (m, c) \in {(0, 0) + [1-9] * [1, 10, 20, 30, 40, 50]}
+ size_distribution: (n, m, c) \in [50, 200, 5] * {(0, 0) + (1, 1) + [3, 5, 7] * [10, 30, 50]}
+ """
+
+ if data_type == "distribution": # focus on TSP100 with gaussian mixture distributions
+ task_set = [(0, 0)] + [(m, c) for m in range(1, 10) for c in [1, 10, 20, 30, 40, 50]]
+ elif data_type == "size": # focus on uniform distribution with different sizes
+ task_set = [(n,) for n in range(min_size, max_size + 1)]
+ elif data_type == "size_distribution":
+ dist_set = [(0, 0), (1, 1)] + [(m, c) for m in [3, 5, 7] for c in [10, 30, 50]]
+ task_set = [(n, m, c) for n in range(50, 201, 5) for (m, c) in dist_set]
+ else:
+ raise NotImplementedError
+
+ print(">> Generating training task set: {} tasks with type {}".format(len(task_set), data_type))
+ print(">> Training task set: {}".format(task_set))
+
+ return task_set
+
diff --git a/rl4co/utils/ops.py b/rl4co/utils/ops.py
index 86dc2a1e..b78821dc 100644
--- a/rl4co/utils/ops.py
+++ b/rl4co/utils/ops.py
@@ -147,7 +147,7 @@ def select_start_nodes(td, env, num_starts):
num_starts: Number of nodes to select. This may be passed when calling the policy directly. See :class:`rl4co.models.AutoregressiveDecoder`
"""
num_loc = env.generator.num_loc if hasattr(env.generator, "num_loc") else 0xFFFFFFFF
- if env.name in ["tsp", "atsp"]:
+ if env.name in ["tsp", "atsp", "flp", "mcp"]:
selected = (
torch.arange(num_starts, device=td.device).repeat_interleave(td.shape[0])
% num_loc
diff --git a/rl4co/utils/test_utils.py b/rl4co/utils/test_utils.py
index 321bf99b..60e2a327 100644
--- a/rl4co/utils/test_utils.py
+++ b/rl4co/utils/test_utils.py
@@ -14,6 +14,8 @@
SMTWTPEnv,
SPCTSPEnv,
TSPEnv,
+ FLPEnv,
+ MCPEnv,
)
@@ -44,6 +46,10 @@ def get_env(name, size):
env = SMTWTPEnv()
elif name == "pdp_ruin_repair":
env = PDPRuinRepairEnv()
+ elif name == "mcp":
+ env = MCPEnv()
+ elif name == "flp":
+ env = FLPEnv()
else:
raise ValueError(f"Unknown env_name: {name}")
diff --git a/tests/test_envs.py b/tests/test_envs.py
index 775a46be..aa83d583 100644
--- a/tests/test_envs.py
+++ b/tests/test_envs.py
@@ -26,6 +26,8 @@
SPCTSPEnv,
SVRPEnv,
TSPEnv,
+ FLPEnv,
+ MCPEnv,
)
from rl4co.utils.decoding import random_policy, rollout
@@ -133,6 +135,13 @@ def test_jssp_lb(env_cls):
assert torch.allclose(td["lbs"], lb_expected)
+@pytest.mark.parametrize("env_cls", [FLPEnv, MCPEnv])
+def test_flp_mcp(env_cls, batch_size=2):
+ env = env_cls()
+ reward, td, actions = rollout(env, env.reset(batch_size=[batch_size]), random_policy)
+ assert reward.shape == (batch_size,)
+
+
def test_scheduling_dataloader():
from tempfile import TemporaryDirectory
diff --git a/tests/test_training.py b/tests/test_training.py
index 4b50f3b0..07d23a25 100644
--- a/tests/test_training.py
+++ b/tests/test_training.py
@@ -17,6 +17,7 @@
DACT,
MDAM,
N2S,
+ POMO,
ActiveSearch,
AttentionModelPolicy,
DeepACO,
@@ -27,9 +28,11 @@
MatNet,
NARGNNPolicy,
NeuOpt,
+ PolyNet,
SymNCO,
)
from rl4co.utils import RL4COTrainer
+from rl4co.utils.meta_trainer import ReptileCallback
# Get env variable MAC_OS_GITHUB_RUNNER
if "MAC_OS_GITHUB_RUNNER" in os.environ:
@@ -128,6 +131,44 @@ def test_mdam():
trainer.test(model)
+def test_pomo_reptile():
+ env = TSPEnv(generator_params=dict(num_loc=20))
+ policy = AttentionModelPolicy(
+ env_name=env.name,
+ embed_dim=128,
+ num_encoder_layers=6,
+ num_heads=8,
+ normalization="instance",
+ use_graph_context=False,
+ )
+ model = POMO(
+ env,
+ policy,
+ batch_size=5,
+ train_data_size=5 * 3,
+ val_data_size=10,
+ test_data_size=10,
+ )
+ meta_callback = ReptileCallback(
+ data_type="size",
+ sch_bar=0.9,
+ num_tasks=2,
+ alpha=0.99,
+ alpha_decay=0.999,
+ min_size=20,
+ max_size=50,
+ )
+ trainer = RL4COTrainer(
+ max_epochs=2,
+ callbacks=[meta_callback],
+ devices=1,
+ accelerator=accelerator,
+ limit_train_batches=3,
+ )
+ trainer.fit(model)
+ trainer.test(model)
+
+
@pytest.mark.parametrize("SearchMethod", [ActiveSearch, EASEmb, EASLay])
def test_search_methods(SearchMethod):
env = TSPEnv(generator_params=dict(num_loc=20))
@@ -159,9 +200,16 @@ def test_nargnn():
@pytest.mark.skipif(
"torch_geometric" not in sys.modules, reason="PyTorch Geometric not installed"
)
+@pytest.mark.skipfif("numba" not in sys.modules, reason="Numba not installed")
def test_deepaco():
env = TSPEnv(generator_params=dict(num_loc=20))
- model = DeepACO(env, train_data_size=10, val_data_size=10, test_data_size=10)
+ model = DeepACO(
+ env,
+ train_data_size=10,
+ val_data_size=10,
+ test_data_size=10,
+ policy_kwargs={"n_ants": 5},
+ )
trainer = RL4COTrainer(
max_epochs=1, gradient_clip_val=1, devices=1, accelerator=accelerator
)
@@ -248,3 +296,17 @@ def test_l2d_ppo(env_cls):
)
trainer.fit(model)
trainer.test(model)
+
+
+def test_polynet():
+ env = TSPEnv(generator_params=dict(num_loc=20))
+ model = PolyNet(
+ env,
+ k=10,
+ train_data_size=10,
+ val_data_size=10,
+ test_data_size=10,
+ )
+ trainer = RL4COTrainer(max_epochs=1, devices=1, accelerator=accelerator)
+ trainer.fit(model)
+ trainer.test(model)
-
-
-
[](https://github.com/psf/black) [](https://join.slack.com/t/rl4co/shared_invite/zt-1ytz2c1v4-0IkQ8NQH4TRXIX8PrRmDhQ)
-[](https://opensource.org/licenses/MIT) 
[](https://pypi.org/project/rl4co)[](https://app.codecov.io/github/ai4co/rl4co/tree/main/rl4co)[](https://github.com/ai4co/rl4co/actions/workflows/tests.yml)
+
+
+
+
+
+
+
+
-
-
-
-where _embeddings_ transfer information from feature space to embedding space.
-
----
-
-
-
-### Constructive Policy Base Classes
-
-```{eval-rst}
-.. automodule:: rl4co.models.common.constructive.base
- :members:
- :undoc-members:
-```
-
-
-
-### Autoregressive Policies Base Classes
-
-```{eval-rst}
-.. automodule:: rl4co.models.common.constructive.autoregressive.encoder
- :members:
- :undoc-members:
-```
-
-```{eval-rst}
-.. automodule:: rl4co.models.common.constructive.autoregressive.decoder
- :members:
- :undoc-members:
-```
-
-```{eval-rst}
-.. automodule:: rl4co.models.common.constructive.autoregressive.policy
- :members:
- :undoc-members:
-```
-
-### Nonautoregressive Policies Base Classes
-
-
-```{eval-rst}
-.. automodule:: rl4co.models.common.constructive.nonautoregressive.encoder
- :members:
- :undoc-members:
-```
-
-```{eval-rst}
-.. automodule:: rl4co.models.common.constructive.nonautoregressive.decoder
- :members:
- :undoc-members:
-```
-
-```{eval-rst}
-.. automodule:: rl4co.models.common.constructive.nonautoregressive.policy
- :members:
- :undoc-members:
-```
diff --git a/docs/_content/api/models/common/improvement.md b/docs/_content/api/models/common/improvement.md
deleted file mode 100644
index 80543b2a..00000000
--- a/docs/_content/api/models/common/improvement.md
+++ /dev/null
@@ -1,3 +0,0 @@
-## Improvement Policies
-
-These methods differ w.r.t. constructive NCO since they can obtain better solutions similarly to how local search algorithms work - they can improve the solutions over time. This is different from decoding strategies or similar in constructive methods since these policies are trained for performing improvement operations.
diff --git a/docs/_content/api/models/common/transductive.md b/docs/_content/api/models/common/transductive.md
deleted file mode 100644
index 6c14c036..00000000
--- a/docs/_content/api/models/common/transductive.md
+++ /dev/null
@@ -1,19 +0,0 @@
-# Transductive Models
-
-
-Transductive models are learning algorithms that optimize on a specific instance. They improve solutions by updating policy parameters $\theta$_, which means that we are running optimization (backprop) **at test time**. Transductive learning can be performed with different policies: for example EAS updates (a part of) AR policies parameters to obtain better solutions, but I guess there are ways (or papers out there I don't know of) that optimize at test time.
-
-
-```{eval-rst}
-.. tip::
- You may refer to the definition of `inductive vs transductive RL  }}){kind=icon}
- {% elif show_source and has_source and sourcename %}
-  }})
- 
"
-
- # TODO: This probably only works when the tutorial list directive is in index.html
- link = self.env.docname + ".html"
-
- if "card_description" in self.options:
- card_description = self.options["card_description"]
- else:
- card_description = ""
-
- if "tags" in self.options:
- tags = self.options["tags"]
- else:
- tags = ""
-
- if "beta" in self.options:
- beta = "Beta"
- else:
- beta = ""
-
- except FileNotFoundError as e:
- print(e)
- return []
- except ValueError as e:
- print(e)
- raise
- return []
-
- card_rst = CARD_TEMPLATE.format(
- header=header,
- image=image,
- link=link,
- card_description=card_description,
- tags=tags,
- beta=beta,
- )
- card_list = StringList(card_rst.split("\n"))
- node = cardnode()
- self.state.nested_parse(card_list, self.content_offset, node)
-
- if not hasattr(self.env, "all_cardnodes"):
- self.env.all_cardnodes = []
- self.env.all_cardnodes.append({"docname": self.env.docname, "node": node})
- return [node]
-
-
-CARD_TEMPLATE = """
-.. raw:: html
-
-
-"""
-
-
-class CustomCalloutItemDirective(Directive):
- option_spec = {
- "header": directives.unchanged,
- "description": directives.unchanged,
- "button_link": directives.unchanged,
- "button_text": directives.unchanged,
- "col_css": directives.unchanged,
- "card_style": directives.unchanged,
- "image_center": directives.unchanged,
- }
-
- def run(self):
- try:
- if "description" in self.options:
- description = self.options["description"]
- else:
- description = ""
-
- if "header" in self.options:
- header = self.options["header"]
- else:
- raise ValueError("header not doc found")
-
- if "button_link" in self.options:
- button_link = self.options["button_link"]
- else:
- button_link = ""
-
- if "button_text" in self.options:
- button_text = self.options["button_text"]
- else:
- button_text = ""
-
- if "col_css" in self.options:
- col_css = self.options["col_css"]
- else:
- col_css = "col-md-6"
-
- if "card_style" in self.options:
- card_style = self.options["card_style"]
- else:
- card_style = "text-container"
-
- if "image_center" in self.options:
- image_center = " }})
-  }})
-  }})
- 
+
+