/Uncertainty-BGNN

Propagation of Aleatoric and Epistemic Uncertainty in Graph Neural Networks

Primary LanguagePython

Uncertainty-BGNN

Propagation of Aleatoric and Epistemic Uncertainty in Graph Neural Networks

This code is provided solely for the purpose of peer review for the ICML 2022 conference.

===================== File specification =========================

  1. src/data/dgl.zip: Modified source files of the package "deep graph library". Unzip for testing.

  2. src/data/Linkprediction_Linkprob.py : Train GNN model for predicting links.

  3. src/data/make_dataset_uncertpropag.py : Preprocess input graph and generate features and labels.

  4. src/data/make_dataset_map.py: Process input graph and generate features and labels from MAP estimate (Baseline).

  5. src/models/Nodeclass_Graphsage.py: Train node classification model.

  6. src/models/Nodeclass_Graphsage_test_mean.py: Test script for mean predictions.

  7. src/models/Nodeclass_Graphsage_test_var.py: Test script for variance estimation.

  8. src/models/Nodeclass_Graphsage_test_MAP.py: Test script for MAP estimation.

  9. utils.py: helping functions.

  10. src/features/build_features.py: helping function for feature extraction

  11. src/visualization/visual.py: helping functions for visualizing features

==================================== end =========================