LGCLD: Layer-adaptive-augmentation-based Graph Contrastive Learning with Feature Decorrelation

This is a PyTorch implementation of LGCLD algorithm, which designs a new graph contrastive learning framework to learn graph-level representations for both unsupervised and semi-supervised graph classification tasks.

Requirements

• python
• pytorch
• pytorch_geometric (pyg)

Note:

This code repository is built on pyg, which is a Python package built for easy implementation of graph neural network model family. Please refer here for how to install and utilize the library.

Datasets

Graph classification benchmarks are publicly available at here.

Run

To run LGCLD, just execute the following command for graph classification task:

python main.py

Reference

[1] Graph Contrastive Learning with Augmentations