This repository contains the implementation of the hypergraph sampling method presented in "Adaptive Sampling for Scalable Hypergraph Neural Networks".
python 3.9 pytorch 1.13.0+cu117 pyg 2.2.0
Create a directory to store hypergraph datasets in a data folder under this repo
1 NVIDIA A100 40GB
Dataset files should be put in the data
folder. Dataset sources are from Semi-supervised Hypergraph Node Classification on Hypergraph Line Expansion, and Datasets, tasks, and training methods for large-scale hypergraph learning
e.g., Run the following to train a GCN classifier with MLP weights on the ModelNet40 dataset:
python main.py --log_wandb=True --dataset ModelNet40 --batch_size 1024 --num_samples 1024 --gflownet_loss LPL --model_type gcn --max_epochs 50 --init_method mlp
e.g., Run the following to train a Graph Transformer classifier on the ModelNet40 dataset:
python main.py --log_wandb=True --dataset ModelNet40 --batch_size 1024 --num_samples 1024 --gflownet_loss LPL --model_type GTransformer --max_epochs 50
Logging on Weights & Biases can be enabled by flag --log_wandb=True
.