First install the graphxai package from the root directory of this project:
pip install -e .
Then you will be able to run test scripts, access features within the package.
All test scripts are in the test_scripts folder in the main repo. Each script is associated with one or two explanation methods, and this is denoted by the naming of the files. For example, test_GBP_MUTAG.py and test_GBP_BA.py both run the Guided Backprop explainer while test_CAM_MUTAG.py and test_CAM_BA.py both run the Class-Activation Mapping (CAM) and Gradient-Class-Activation Mapping (Grad-CAM). The suffix ..._MUTAG.py denotes that the script runs an explanation on the MUTAG dataset and ..._BA.py denotes the script running an explanation on a custom version of the synthetic BA houses dataset, as is described in GNNExplainer paper.
To run the BA scripts, you'll simply run the script directly. For example, to run test_CAM_BA.py, the test script for CAM and Grad-CAM on the BA houses dataset, you'll run:
>>> python3 test_CAM_BA.py
To run the MUTAG scripts, you'll need to provide an additional command-line argument which corresponds to the index of the molecule for which you want to show an explanation. This is mainly for consistency across multiple runs of these scripts. For example, to run test_GBP_MUTAG.py, the test script for Guided Backprop on the MUTAG dataset, with the molecule at index 6, you'll run:
>>> python3 test_GBP_MUTAG.py 6
Note that all of these scripts will both provide output through stdout as well as show an image through matplotlib.
ShapeGraph 1 is a large synthetic graph that has many desirable properties for evaluating GNN explanations. You load it in the following way:
from graphxai.datasets import load_ShapeGraph
SG = load_ShapeGraph(number=1)
# SG now holds a fully loaded and verified ShapeGraph object.