This repository is the official implementation of "CAT-WALK: Inductive Hypergraph Learning via Set Walks".
Ali Behrouz, Farnoosh Hashemi, Sadaf Sadeghian, Margo Seltzer
Temporal hypergraphs provide a powerful paradigm for modeling time-dependent, higher-order interactions in complex systems. Representation learning for hypergraphs is essential for extracting patterns of the higher-order interactions that are critically important in real-world problems in social network analysis, neuroscience, finance, etc. However, existing methods are typically designed only for specific tasks or static hypergraphs. We present CAT-Walk, an inductive method that learns the underlying dynamic laws that govern the temporal and structural processes underlying a temporal hypergraph. CAT-Walk introduces a temporal, higher-order walk on hypergraphs, SetWalk, that extracts higher-order causal patterns. CAT-Walk uses a novel adaptive and permutation invariant pooling strategy, SetMixer, along with a set-based anonymization process that hides the identity of hyperedges. Finally, we present a simple yet effective neural network model to encode hyperedges. Our evaluation on 10 hypergraph benchmark datasets shows that CAT-Walk attains outstanding performance on temporal hyperedge prediction benchmarks in both inductive and transductive settings. It also shows competitive performance with state-of-the-art methods for node classification.
python >= 3.7,PyTorch >= 1.4, please refer to their official websites for installation details.- Other dependencies:
pandas==0.24.2
tqdm==4.41.1
numpy==1.16.4
scikit_learn==0.22.1
numba==0.51.2
We use 10 publicly available datasets that can be found here. To use them please put the data under HG_Data floder.
usage: Interface for CATWALK: Inductive Dynamic Representation Learning for Link Prediction on Temporal Hyper Graphs
[-h]
[-d {NDC-classes,contact-primary-school,contact-high-school,email-Enron,email-Eu,congress-bills,tags-math-sx,threads-ask-ubuntu,NDC-substances,tags-ask-ubuntu}]
[-m {t,i}] [--pretrained {True,False}] [--n_degree [N_DEGREE [N_DEGREE ...]]] [--n_layer N_LAYER]
[--bias BIAS] [--pos_dim POS_DIM] [--pos_sample {multinomial,binary}]
[--he_encode_hid_dim HE_ENCODE_HID_DIM] [--he_encode_out_dim HE_ENCODE_OUT_DIM]
[--walk_encode_time_dim WALK_ENCODE_TIME_DIM] [--walk_encode_hid_dim WALK_ENCODE_HID_DIM]
[--walk_encode_out_dim WALK_ENCODE_OUT_DIM] [--src_he_encode_hid_dim SRC_HE_ENCODE_HID_DIM]
[--src_he_encode_out_dim SRC_HE_ENCODE_OUT_DIM] [--task_layer1_out_dim TASK_LAYER1_OUT_DIM]
[--max_he_size MAX_HE_SIZE] [--walk_agg {set_node_gran,mean_he_gran,mean_node_gran}]
[--sampled_he_per_node SAMPLED_HE_PER_NODE] [--n_epoch N_EPOCH] [--bs BS] [--lr LR]
[--drop_out DROP_OUT] [--tolerance TOLERANCE] [--seed SEED] [--gpu GPU] [--cpu_cores CPU_CORES]
[--verbosity VERBOSITY] [--partial_e_num PARTIAL_E_NUM]optional arguments:
-h, --help show this help message and exit
-d {NDC-classes,contact-primary-school,contact-high-school,email-Enron,email-Eu,congress-bills,tags-math-sx,threads-ask-ubuntu,NDC-substances,tags-ask-ubuntu}, --data {NDC-classes,contact-primary-school,contact-high-school,email-Enron,email-Eu,congress-bills,tags-math-sx,threads-ask-ubuntu,NDC-substances,tags-ask-ubuntu}
data sources to use, try email-Enron or NDC-classes
-m {t,i}, --mode {t,i}
transductive (t) or inductive (i)
--pretrained {True,False}
use pretrained models or not
--n_degree [N_DEGREE [N_DEGREE ...]]
a list of neighbor sampling numbers for different hops, when only a single element is
input n_layer will be activated
--n_layer N_LAYER number of network layers
--bias BIAS the hyperparameter alpha controlling sampling preference in recent time, default to 0
which is uniform sampling
--pos_dim POS_DIM dimension of the positional embedding
--pos_sample {multinomial,binary}
two equivalent sampling method with empirically different running time
--he_encode_hid_dim HE_ENCODE_HID_DIM
dimension of the hidden embedding of hypergraph encoder
--he_encode_out_dim HE_ENCODE_OUT_DIM
dimension of the hypergraph embeddding
--walk_encode_time_dim WALK_ENCODE_TIME_DIM
dimension of the time embeddding in walk encoder
--walk_encode_hid_dim WALK_ENCODE_HID_DIM
dimension of the hidden embeddding in walk encoder
--walk_encode_out_dim WALK_ENCODE_OUT_DIM
dimension of the walk embeddding
--src_he_encode_hid_dim SRC_HE_ENCODE_HID_DIM
dimension of the hidden embedding of source hypergraph encoder
--src_he_encode_out_dim SRC_HE_ENCODE_OUT_DIM
dimension of the source hypergraph embeddding
--task_layer1_out_dim TASK_LAYER1_OUT_DIM
dimension of the output of the first task layer
--max_he_size MAX_HE_SIZE
maximum size (number of nodes) of a hypergraph
--walk_agg {set_node_gran,mean_he_gran,mean_node_gran}
aggregation method of walk encoding to find source hyperedge encoding
--sampled_he_per_node SAMPLED_HE_PER_NODE
number of sampled he per node in node classification
--n_epoch N_EPOCH number of epochs
--bs BS batch_size
--lr LR learning rate
--drop_out DROP_OUT dropout probability for all dropout layers
--tolerance TOLERANCE
tolerated marginal improvement for early stopper
--seed SEED random seed for all randomized algorithms
--gpu GPU which gpu to use
--cpu_cores CPU_CORES
number of cpu_cores used for position encoding
--verbosity VERBOSITY
verbosity of the program output
--partial_e_num PARTIAL_E_NUM
number of hyperedges to load from the dataset(when loading the dataset partially)
We have utilized the code from here as the foundation of our implementation, customizing it extensively to align with our specific needs. Thanks to the authors for sharing their code.
catwalk,
title={CAT-Walk: Inductive Hypergraph Learning via Set Walks},
author={Ali Behrouz, Farnoosh Hashemi, Sadaf Sadeghian, Margo Seltzer},
booktitle={Thirty-seventh Conference on Neural Information Processing Systems},
year={2023},
url={https://openreview.net/forum?id=QG4nJBNEar}
}