RexYing/gnn-model-explainer
Gnn Model Explainer
This toolkit serves as a framework for interpreting the decision-making processes of graph neural networks. It functions as a library for analyzing how these models process complex network data, providing methods to identify the specific node attributes and structural patterns that influence predictive outcomes.
The project distinguishes itself by employing mask-optimized subgraph extraction and gradient-based attribution mapping to isolate the minimal components of a graph that preserve a model's original prediction. By separating graph processing layers from explanation logic, the architecture allows for consistent analysis across various types of graph neural networks, including convolutional and attention-based models.
Beyond core interpretation, the library supports the training of graph models on synthetic and real-world datasets and facilitates machine learning model debugging. It includes capabilities for visualizing model explanations through interactive notebooks, enabling the rendering of attribution heatmaps and complex network structures to assist in the identification of influential features and subgraphs.
Features
- Graph Neural Network Explainers - Identifies influential subgraphs and node features that drive predictive outcomes in graph neural network models.
- Model Interpretability Tools - Determine which specific node attributes or data points have the greatest impact on the decision-making process of a model when making predictions based on graph data.
- Graph - Analyzes how graph-based machine learning models make predictions by identifying influential nodes and structural patterns.
- Mask-Optimized - Learns binary masks over graph components to isolate the minimal structural subset that preserves the original model prediction.
- Feature Importance Attribution - Quantifies the relative contribution of individual input variables to the predictions of machine learning models.
- Attribution Gradients - Computes feature importance by backpropagating model outputs through the computational graph to input gradients.
- Graph Data Analysis Libraries - Provides a collection of routines for training graph convolutional and attention models on network datasets.
- Graph Message Passing Frameworks - Implements neural message passing primitives to learn representations of nodes and edges in graph structures.
- Graph Neural Network Training Routines - Provides routines for training graph convolutional and attention models on synthetic and real-world datasets.
- Machine Learning Debugging - Investigates the internal logic of complex neural networks to identify biases or errors in graph data processing.
- Model Explanation Visualizations - Renders generated masks and model insights through interactive notebooks to facilitate detailed analysis of model interpretations.
- Network Graph Visualization - Renders complex network structures and model insights into clear visual formats to help researchers understand data flow.
- Subgraph Matching - Isolates specific structural components within a graph that contribute most significantly to a model prediction.
