direpack: A Python 3 package for state-of-the-art statistical dimension reduction methods

TL;DR

The direpack package consolidates modern statistical dimension reduction techniques into a single, scikit-learn compatible Python package, facilitating advanced data preprocessing and regression methods within machine learning workflows.

Contribution

It introduces a comprehensive Python package that integrates various state-of-the-art dimension reduction methods with utilities for preprocessing and regression, all compatible with scikit-learn.

Findings

Provides a unified, easy-to-use interface for multiple dimension reduction techniques.

Includes robust and classical preprocessing utilities for improved data analysis.

Enables seamless integration into machine learning pipelines.

Abstract

The direpack package aims to establish a set of modern statistical dimension reduction techniques into the Python universe as a single, consistent package. The dimension reduction methods included resort into three categories: projection pursuit based dimension reduction, sufficient dimension reduction, and robust M estimators for dimension reduction. As a corollary, regularized regression estimators based on these reduced dimension spaces are provided as well, ranging from classical principal component regression up to sparse partial robust M regression. The package also contains a set of classical and robust pre-processing utilities, including generalized spatial signs, as well as dedicated plotting functionality and cross-validation utilities. Finally, direpack has been written consistent with the scikit-learn API, such that the estimators can flawlessly be included into (statistical and/or machine) learning pipelines in that framework.