Meta Learning for Everyone: Developing Few-shot Learning Models and Fast Reinforcement Learning Agents using PyTorch
This repository is a repository for the book "Meta-Learning for Everyone".
This repository is implemented and verified on python 3.8.15.
First, install Anaconda from the link below.
Second, follow the commands below to create a new python environment and activate the created environment.
(base) $ conda create -y -n meta python=3.8.8
(base) $ conda activate meta
(meta) $ conda env list
Next, after cloning this repository, run the following command to install the required packages.
MacOS & Linux user
# User
(meta) $ make init
# Developer
(meta) $ make init-dev
Windows user
# User
(meta) $ "./scripts/window-init.bat"
Meta-SL
For Meta-SL, move to each algorithm folder, run the algorithms using jupyter notebook
, and check the results.
(meta) $ jupyter notebook
If you're using Colab, please refer to the Installation of Torchmeta in Colab guide to install Torchmeta.
Meta-RL
For Meta-RL, move to each algorithm folder and run the commands below.
# RL^2
(meta) $ python rl2_trainer.py
# MAML
(meta) $ python maml_trainer.py
# PEARL
(meta) $ python pearl_trainer.py
In the case of Meta-RL, please run the Tensorboard command below to check the results of the meta-training and meta-testing you executed.
(meta) $ tensorboard --logdir=./results
Thanks goes to these wonderful people (emoji key):
Dongmin Lee 💻 📖 |
Seunghyun Lee 💻 📖 |
Luna Jang 💻 |
Seungjae Ryan Lee 💻 |
This project follows the all-contributors specification. Contributions of any kind welcome!