causalOT: Optimal transport methods for causal inference

This R package implements the methods described in Optimal transport methods for causal inference.

Installation

This package can be installed in a few ways.

1. devtools

Using the remotes package in R, one can install the package with

remotes::install_github("ericdunipace/causalOT")

2. download and install

After downloading the git package using git clone or by downloading the .zip file from the button above (Code -> Download Zip) and unzipping, you can install the package with

devtools::install("path/to/causalOT")

3. CRAN

A stable version of this package is available on CRAN, but usually this GitHub will have the latest version.

Additional required software

If using Causal Optimal Transport based on the Sinkhorn divergence, you will need to install Python 3.x and the geomloss package, which depends on numpy, scipy, torch, and pykeops.

Usage

The functions in the package are built to construct weights to make distributions more same and estimate causal effects. The primary method we recommend is by using optimal transport weights which balance distributions by design. For more information about using this package, see the vignette “Using causalOT”.

Reproducing the paper

In the folder inst/Reproduce you can find code and an RMarkdown file to reproduce the figures present in the paper.

Package author

Eric Dunipace

License

This package is licensed under GPL 3.0.