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 =========================
-
src/data/dgl.zip: Modified source files of the package "deep graph library". Unzip for testing.
-
src/data/Linkprediction_Linkprob.py : Train GNN model for predicting links.
-
src/data/make_dataset_uncertpropag.py : Preprocess input graph and generate features and labels.
-
src/data/make_dataset_map.py: Process input graph and generate features and labels from MAP estimate (Baseline).
-
src/models/Nodeclass_Graphsage.py: Train node classification model.
-
src/models/Nodeclass_Graphsage_test_mean.py: Test script for mean predictions.
-
src/models/Nodeclass_Graphsage_test_var.py: Test script for variance estimation.
-
src/models/Nodeclass_Graphsage_test_MAP.py: Test script for MAP estimation.
-
utils.py: helping functions.
-
src/features/build_features.py: helping function for feature extraction
-
src/visualization/visual.py: helping functions for visualizing features
==================================== end =========================