Causal Discovery with Deductive Reasoning: One Less Problem

This python package is for the paper titled Causal Discovery with Deductive Reasoning: One Less Problem (UAI 2024).

Since submission, we modified many parts of our codebase implementing additional functionalities. Regardless, we organized python3 files into folders such as scripts, experiments, and algos (algorithms) where scripts folder is separated from module cddd in source folder src. Due to some code changes, we do not exactly reproduce the results since some of the random seeds are changed. But there will be no problem observing trends similar to the tables and figures reported in the paper.

Preparation

At this moment, we only support Unix-based systems (we hard-coded path separators as /). You may place CD_DD in your home directory. (Otherwise, you may need to change some variables in python scripts later.) Please use python 3.11 since we did not test other versions. You may install python packages necessary to run experiments by first checking requirements.txt (you may ignore version constraints for the packages listed in the requirements.txt.)

pip3 install -r requirements.txt

Install the Package:

Now, move to the source directory in CD_DD and install the module with pip. We will assume that you downloaded CD_DD in your home directory.

cd ~/CD_DD/src

pip3 install .

Running Scripts

To run the algorithms, you will need to 1) generate data, 2) conduct experiments, and 3) draw results. Those files are in scripts folder, scripts

cd ~/CD_DD/scripts

You can either execute bash run.sh to run the complete process or run each of the following scripts individually. Here are the details for each script:

Generating Data:

sample_data.py script will create two directories Ground_truth and Sampled_datasets under ~/CD_DD/data and generate DAGs and corresponding samples used in the paper.
Please change WORKING_DIR variable in the script properly. Currently, WORKING_DIR = '~/CD_DD'

python3 sample_data.py

It will take about 2 minutes. Note that there will be warning messages about probabilities not being sum to 1.0. You can ignore those warnings. Also it may take (less than) 1GB of storage.

Conducting Experiments:

conduct_experiments.py will run multiple experiments simultaneously. If you don't need to reproduce results similar to the paper, you may select some of the datasets and parameters for algorithms. ~/CD_DD/results will be created and csv files will be generated. You can also control parallelism by changing values for n_jobs (the number of CPU cores to utilize.) Again, please change WORKING_DIR variable in the script properly.

python3 conduct_experiments.py

Note that full experiments may take about 4 hours utilizing 128 CPUs.

Drawing Plots:

draw_results.py script will draw plots (and save them to results folder) based on the csv files saved in results after conduct_experiments.py. Note that the code may not output plots properly if some of experiments are not conducted. Sorry for the inconvenience.

python3 draw_results.py

Citation

If you find our work useful for your research, please consider citing our paper.

@inproceedings{kim2024causal,
  title={Causal Discovery with Deductive Reasoning: One Less Problem},
  author={Kim, Jonghwan and Hwang, Inwoo and Lee, Sanghack},
  booktitle={Proceedings of the Fortieth Conference on Uncertainty in Artificial Intelligence},
  year={2024},
  publisher={PMLR}
}