KCRL: Prior Knowledge Based Causal Discovery With Reinforcement Learning

This is the public repository of code implementation for KCRL: A Prior Knowledge Based Causal Discovery Framework With Reinforcement Learning accepted for inclusion in the MLHC 2022 research track proceedings.

Requirements

The code is tested with the following configuration:

Python=3.6.10, numpy=1.18.1, pandas=1.0.1, scikit-learn=0.22.1, scipy=1.4.1, tensorflow=1.13.1, networkx, pyyaml=5.3, pytz=2019.3, matplotlib=3.1.3

Instructions

Test KCRL with any dataset:

python kcrl_demo.py

Detailed instructions:

1. Set the initial parameters in the kcrl_demo.py file. (Detailed comments available inside the file.)
2. To use prior knowledge, provide the existing edges (0 or 1) in the prior knowledge matrix in kcrl_demo.py file. (Detailed comments available inside the file.)

Acknowledgement

Our code has been benefited from the following existing works. We are really thankful to the corresponding authors.

DAGs With NOTEARS https://github.com/xunzheng/notears

Causal Discovery With Reinforcement Learning https://github.com/huawei-noah/trustworthyAI.git

gCastle package https://github.com/huawei-noah/trustworthyAI/tree/master/gcastle

Neural Combinatorial Optimization with RL https://github.com/MichelDeudon/neural-combinatorial-optimization-rl-tensorflow