llja0112 / counterfactual-gp

Counterfactual Gaussian Processes

counterfactual-gp

version number: 0.0.1

author: Guangyi Zhang

Overview

Counterfactual Gaussian Processes

Maximize the following MPP (marked point process) to build a counterfactual model.

• The event model, i.e., the point process, is omitted since it does not affect the outcome.
• The action model has separate parameters from the outcome model.
• The action model is time-invariant, i.e., playing a similar role as the mixture coefficients in mixture model of GPs.
• So far the action model is also binary and individual-invariant, which, however, can be easily extended.

Reference:

• Schulam, Peter, and Suchi Saria. "Reliable decision support using counterfactual models." Advances in Neural Information Processing Systems. 2017.

Usage

TBD

Dev

Build simulated dataset from Jupyter Notebooks.

To run tests:

$ py.test

Contributing

TBD

Example

TBD