IVAPS: Algorithm is Experiment Documentation

Overview

The IVaps package is an implementation of the treatment effect estimation method proposed in [1]. The key to the method is an object which captures the average probability of an algorithm \(A\) recommending treatment for an individual with observable covariates \(x\) in a shrinking neighborhood around \(x\). This object is dubbed the Approximate Propensity Score (APS) by [1]. The package provides functions for APS estimation and treatment effect estimation, and is designed to be flexible to the researcher’s specific treatment interface.

Supported ML Frameworks

The APS estimation function estimate_aps_onnx only accepts models in the ONNX framework in order to maintain the generalized implementation. The module provides an ONNX conversion function convert_to_onnx that currently supports conversion from the following frameworks:

• Sklearn

• Pytorch

• Tensorflow

• LightGBM

• Keras

• XGBoost (experimental)

• CatBoost (experimental)

• CoreML (experimental)

• LibSVM (experimental)

• SparkML (experimental)

For conversion functions for other frameworks, please refer to the onnxmltools repository. Please note that use of these functions requires that the relevant framework packages are installed.

License

This project is licensed under the Apache 2.0 License - see the LICENSE file for details.

Introduction:

• About
  • ML as a Data Production Service
  • MLisNE Pipeline
• IVaps Framework Overview
  • Setup
  • APS
  • Identification
  • Examples

Getting Started:

• Installation
• Quickstart
  • APS Estimation
  • IV Estimation
  • Counterfactual Estimation
  • Model Conversion
• Examples

API Reference:

• IVAPS API Reference
  • Modules

Others:

• References
• Github

Indices and tables

• Index

• Module Index

• Search Page