Welcome to CFL

Causal Feature Learning (CFL) is an unsupervised algorithm designed to construct macro-variables from low-level data, preserving the causal relationships present in the data.

Getting Started

• Install CFL
• What is Causal Feature Learning?
• CFL Package In-Depth Tutorial
• CFL Package Quick Start

In-Depth Feature Tutorials

• CondDensityEstimator Hyperparameter Tuning with Optuna
• CauseClusterer and EffectClusterer Hyperparameter Tuning
• Built-in Macrostate Visualizations
• How to add new CondDensityEstimators, CauseClusterers, and EffectClusterers to CFL

Dataset Applications

• Applying CFL to the Visual Bars Dataset
• Applying CFL to the El Niño Dataset

Contribute to CFL

• Contribute to CFL

API Reference

User API

More Info

• Background: Visual Bars Data
• Some Notes on Conditional Density Estimators (CDEs)
• Some Notes on Clustering
• How to Use Data Version Control (DVC)

Contributors

• Jenna Kahn & Iman Wahle [first authors; name order chosen randomly]

• Krzysztof Chalupka

• Daniel Israel

• Patrick Burauel

• Pietro Perona

• Frederick Eberhardt

Jenna Kahn and Iman Wahle designed the software and wrote the code in this repository. Daniel Israel wrote the MNIST example notebook and contributed feedback about the code.

Krzysztof Chalupka, Pietro Perona and Frederick Eberhardt developed the original theory for CFL. Krzysztof also wrote the original code upon which this software is based.

Code development benefitted from regular discussions with Patrick Burauel.

License and Citations

CFL is released under a BSD-like license for non-commercial use only. If you use CFL in published research work, we encourage you to cite this repository:

Causal Feature Learning (2022). https://github.com/eberharf/cfl

or use the BibTex reference:

@misc{cfl2022,
title = "Causal Feature Learning",
year = "2022",
publisher = "GitHub",
url = "https://github.com/eberharf/cfl"}