pyg-captum-shap 0.1.2

A robust SHAP explainer wrapper for PyTorch Geometric models.

PyG-SHAP

A robust wrapper bridging the Euclidean sampling mechanisms of Captum with the non-Euclidean batching of PyTorch Geometric (PyG). Designed specifically for molecular QSAR modelling. This library is built on top of the Captum library by PyTorch.

This library provides a Dictionary-Shielded Wrapper that:

1. Intercepts Captum's corrupted sampling tensors.
2. Safely reconstructs the block-diagonal graph batches on-the-fly.
3. Allows for seamless SHAP and Integrated Gradients attribution on complex PyG architectures (GAT, GCN, Transformers).

Installation

pip install pyg-captum-shap

Quick Start

from pyg_captum_shap import compute_gat_shap_values

# Extract SHAP values for a specific molecule and task
node_attributions = compute_gat_shap_values(
    model=your_trained_model,
    target_graph=molecule_graph_data,
    target_task=0
)

# node_attributions now contains the importance score for every atom in the graph

License

Distributed under the MIT License.