Variational Causal Inference

This repository is the official implementation of Variational Causal Inference.

Installation

1. Create Conda Environment

conda config --append channels conda-forge
conda create -n vci-env --file requirements.txt
conda activate vci-env

2. Install Learning Libraries

• Pytorch 1.13.1

  * make sure to install the right versions for your toolkit

Data

Visit our resource site, download the contents of main/datasets into datasets. To see how to process your own dataset, download the contents of data-prep/datasets into datasets and follow the examples. A clean example of data preparation can be found in SciplexPrep.ipynb. For an example of data preparation on a messier dataset with thorough analysis and visualizations, see MarsonPrep.ipynb.

In summary, the preparation procedure includes:

• Quality Control

• Normalization

• Gene Selection

• Field Specification

• (Optional) Dataset Split

• (Optional) Out-of-distribution Selection

• (Optional) Differentially-expressed Gene Ranking

Run

Once the environment is set up and the data are prepared, the function call to train the model is:

./main.sh &

A list of flags may be found in main.sh and main.py for experimentation with different network parameters. The run log and models are saved under *artifact_path*/saves, and the tensorboard log is saved under *artifact_path*/runs.

License

Contributions are welcome! All content here is licensed under the MIT license.