GNN Explainability Framework

Node Classification Tasks

Explainer	Paper
Distance	Shortest Path Distance Approximation using Deep learning Techniques
PageRank	The PageRank Citation Ranking: Bringing Order to the Web
SA	Explainability Techniques for Graph Convolutional Networks.
Grad-CAM	Explainability Methods for Graph Convolutional Neural Networks.
Integrated Gradients	Axiomatic Attribution for Deep Networks
GNNExplainer	GNNExplainer: Generating Explanations for Graph Neural Networks
PGExplainer	Parameterized Explainer for Graph Neural Network
SubgraphX	On Explainability of Graph Neural Networks via Subgraph Exploration
PGM-Explainer	PGM-Explainer: Probabilistic Graphical Model Explanations for Graph Neural Networks

Installation

Requirements

CPU or NVIDIA GPU, Linux, Python 3.7
PyTorch >= 1.5.0, other packages

Pytorch Geometric. Official Download.

# We use TORCH version 1.6.0
CUDA=cu111
TORCH=1.9.1
pip install torch-scatter -f https://pytorch-geometric.com/whl/torch-${TORCH}+${CUDA}.html
pip install torch-sparse -f https://pytorch-geometric.com/whl/torch-${TORCH}+${CUDA}.html
pip install torch-cluster -f https://pytorch-geometric.com/whl/torch-${TORCH}+${CUDA}.html
pip install torch-spline-conv -f https://pytorch-geometric.com/whl/torch-${TORCH}+${CUDA}.html
pip install torch-geometric==2.0.3

Other packages

pip install tqdm matplotlib argparse json jupyterlab notebook pgmpy captum

Datasets

The processed raw data for datasets ba_house, ba_community, ba_grid, tree_cycle, tree_grid, ba_bottle is available in the data/syn folder.
The processed raw data for datasets cora, citeseer, pubmed, cornell, texas, wisconsin, chameleon, squirrel, actor will be automatically downloaded when training models.

Python code map

.
├── dataset
│   ├── __init__.py
│   ├── data_utils.py
│   ├── gen_mutag.py
│   ├── gen_real.py
│   ├── gen_syn.py
│   ├── mutag_utils.py
│   └── syn_utils
│       ├── featgen.py
│       ├── gengraph.py
│       ├── gengroundtruth.py
│       └── synthetic_structsim.py
├── evaluate
│   ├── __init__.py
│   ├── accuracy.py
│   ├── fidelity.py
│   └── mask_utils.py
├── explainer
│   ├── __init__.py
│   ├── genmask.py
│   ├── gnnexplainer.py
│   ├── graph_explainer.py
│   ├── node_explainer.py
│   ├── pgmexplainer.py
│   ├── pgexplainer.py
│   ├── shapley.py
│   └── subgraphx.py
├── gnn
│   ├── __init__.py
│   ├── eval.py
│   ├── model.py
│   └── train.py
├── main.py
└── utils
    ├── __init__.py
    ├── gen_utils.py
    ├── graph_utils.py
    ├── io_utils.py
    ├── math_utils.py
    ├── parser_utils.py
    └── plot_utils.py

Node Classification

python3 code/main.py --dataset [dataset-name] --explain_graph False --explainer_name [explainer_name]

dataset-name:
- synthetic: ba_house, ba_grid, tree_cycle, tree_grid, ba_bottle
- real-world: cora, pubmed, citeseer, facebook, chameleon, squirrel, texas, wisconsin, cornell, actor
explainer_name: random, pagerank, distance, sa, ig, gradcam, occlusion, basic_gnnexplainer, gnnexplainer, subgraphx, pgmexplainer, pgexplainer

Note that gradcam is only available for synthetic datasets.

Mask transformation

To compare the methods, we adopt separately three strategies to cut off the masks:

Sparsity
Threshold
Topk

This can be changed by changing the --strategy parameter. Choices are [topk, sparsity,threshold]. The default strategy is topk. You adjust the level of transformation with the --params_list parameter. Here, you define the list of transformation values. Default list is "5,10"

Jupyter Notebook

The default visualizations are provided in notebook/GNN-Explainer-Viz.ipynb.

Note: For an interactive version, you must enable ipywidgets
jupyter nbextension enable --py widgetsnbextension

Tuning the mask sparsity/threshold/top-k values.

Included experiments

Name	`EXPERIMENT_NAME`	Description
Barabasi-House	`ba_house`	Random BA graph with House attachments.
Barabasi-Grid	`ba_grid`	Random BA graph with grid attachments.
Tree-Cycle	`tree_cycle`	Random Tree with cycle attachments.
Tree-Grid	`tree_grid`	Random Tree with grid attachments.
Barabasi-Bottle	`ba_bottle`	Random BA graph with bottle attachments.
MUTAG	`mutag`	Mutagenecity Predicting the mutagenicity of molecules (source).
Cora	`cora`	Citation network (source).
Pubmed	`pubmed`	PubMed network (source).
Citeseer	`citeseer`	Citeseer network (source).
FacebookPagePage	`facebook`	Facebook Page-Page network dataset
Chameleon	`chameleon`	Wikipedia dataset
Squirrel	`squirrel`	Wikipedia dataset
Texas	`texas`	WebKB dataset
Wisconsin	`wisconsin`	WebKB dataset
Cornell	`cornell`	WebKB dataset
Actor	`actor`	Film-director-actor-writer network (Actor)

Using the explainer on other models

A graph convolutional model is provided. This repo is still being actively developed to support other GNN models in the future.

Citation

Please cite our paper if you find the repository useful.

zhangce/GraphFramEx