All notable changes to this project will be documented in this file.
The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.
- Refactor code, supporting only inductive setting
- Removed unused methods/modules
- Add install instructions to README
- Provide results/reproduction steps in README
- Update to latest version of torch geometric (2.0.1)
- Use datasets and utilities from torch geometric
- Refactor datasets creation
- ** Note: Datasets are not the same as in the original paper
- Graphs are undirected (twice the number of edges)
- Different train/validation/test splits
- Different method for negative sampling, negative sampling fraction
- ** Note: Datasets are not the same as in the original paper
- Support both inductive and transductive validation
- For transductive validation, the Validation set subgraph is part of training set. For inductive validation, these nodes/edges are unseen during training.
- All inference on the test set is inductive
- We set lambda_0 -> 0 so that only node attributes are taken into effect