daehoum1 / pcfi

Confidence-Based Feature Imputation for Graphs with Partially Known Features (ICLR 2023)

PCFI (ICLR2023)

This repository is the official PyTorch implementation of "Confidence-Based Feature Imputation for Graphs with Partially Known Features" [Paper]. The codes are built on Feature Propagation.

Requirements

python >= 3.9
torch == 1.10.2
pyg == 2.0.3

To run the code

Semi-supervised node classification

python run_node.py

Link prediction

python run_link.py

Citation

@inproceedings{umconfidence,
  title={Confidence-Based Feature Imputation for Graphs with Partially Known Features},
  author={Um, Daeho and Park, Jiwoong and Park, Seulki and young Choi, Jin},
  booktitle={The Eleventh International Conference on Learning Representations}
}