API Reference

The BRIDGE library consists of several core modules and subpackages that provide functionality for graph rewiring, sensitivity analysis, and experiment utilities.

Main Components

models: GCN and SelectiveGCN implementations
rewiring: Graph rewiring operations and pipelines
training: Training loops for graph neural networks
utils: Utility functions for graphs and matrices
datasets: Synthetic graph dataset generation
optimization: Hyperparameter optimization for GNNs
sensitivity: Sensitivity and SNR analysis tools

Core Modules

bridge.models

The models package provides implementations of Graph Convolutional Networks (GCNs) and their variants.

GCN: Standard Graph Convolutional Network implementation
HPGraphConv: High-Pass Graph Convolution layer
SelectiveGCN: GCN that can selectively operate on different graph structures

bridge.rewiring

The rewiring package contains functions for rewiring graph structures to optimize MPNN performance.

run_bridge_pipeline: Main pipeline for running the BRIDGE rewiring process
run_bridge_experiment: Function for running multiple rewiring trials across data splits
run_iterative_bridge_pipeline: Iterative rewiring pipeline
run_iterative_bridge_experiment: Repeated trials for iterative rewiring
create_rewired_graph: Low-level function to create a rewired version of a graph

bridge.training

The training package provides functions for training and evaluating graph neural networks.

train: Train a GNN with early stopping
train_selective: Train a selective GNN on multiple graph versions

bridge.utils

The utils package includes utility functions for working with graphs and matrices.

local_homophily: Compute local homophily for each node
matrix_utils: Utilities for working with matrices
graph_utils: Utilities for working with graphs

bridge.datasets

The datasets package provides functionality for creating and working with graph datasets.

SyntheticGraphDataset: Generate synthetic graphs with controllable properties

bridge.optimization

The optimization package provides objective functions for hyperparameter optimization with Optuna.

objective_gcn: Objective function for optimizing base GCN hyperparameters
objective_rewiring: Objective function for optimizing rewiring parameters
objective_iterative_rewiring: Objective for iterative rewiring

bridge.sensitivity

The sensitivity package offers tools for sensitivity and SNR analysis of graph neural networks.

SNR Estimation: Estimate the SNR of GNN outputs
Sensitivity Analysis: Compute sensitivity measures for GNNs
Feature Generation: Generate features with controlled covariance structure