/DR-VIDAL-AMIA-22

Official repository of DR-VIDAL - accepted in AMIA' 22 (Oral)

Primary LanguagePythonMIT LicenseMIT

Official implementation of DR-VIDAL - Doubly Robust Variational Information(Theoretic) Adversarial Learning for the Estimation of Treatment Effects and Counterfactuals (AMIA, Symoissium, 2022)

Paper | Arxiv

Keywords

causal AI, biomedical informatics, generative adversarial networks, variational inference, information theory, doubly robust

Supplementary Material for the paper

The supplementary material with proofs and additional results will be found at: here.

Presentation slides

The presentation slides will be found at:

  1. PPT
  2. PDF

Presentation video

Click this link.

Packages

All the packages are inluded in environment.yml file

Overview

Requirements and versions

  • pytorch - 1.3.1
  • numpy - 1.17.2
  • pandas - 0.25.1
  • scikit - 0.21.3
  • matplotlib - 3.1.1
  • python - 3.8

Dependencies

python 3.8

pytorch 1.3.1

How to run

First go the folder DR_Info_CFR by the command cd DR_Info_CFR and then do the following for each of the 3 datasets:

  • IHDP:

Command to reproduce the experiments mentioned in the paper for IHDP dataset:

cd IHDP

python3 main_IHDP.py

  • Jobs:

Command to reproduce the experiments mentioned in the paper for Jobs dataset:

cd Jobs

python3 main_Jobs.py

  • Twins:

Command to reproduce the experiments mentioned in the paper for Twins dataset:

cd Twins

python3 main_Twins.py

Hyperparameters

  • IHDP: IHDP/Constants.py

  • Jobs: Jobs/Constants.py

  • Twins: Twins/Constants.py

Results

Cite

@inproceedings{ghosh2021dr,
  title={DR-VIDAL-Doubly Robust Variational Information-theoretic Deep Adversarial Learning for Counterfactual Prediction and Treatment Effect Estimation on Real World Data},
  author={Ghosh, Shantanu and Feng, Zheng and Bian, Jiang and Butler, Kevin and Prosperi, Mattia},
  booktitle={AMIA Annual Symposium Proceedings},
  volume={2022},
  pages={485},
  year={2022},
  organization={American Medical Informatics Association}
}

License & copyright

Licensed under the MIT License

Copyright (c) DISL, 2021