Welcome to the repository for BasisDeVAE: Interpretable Simultaneous Dimensionality Reduction and Feature-Level Clustering with Derivative-Based Variational Autoencoders, presented at ICML 2021.

The code provided here builds on the original implementation of BasisVAE by Kaspar Märtens in PyTorch, to be found at https://github.com/kasparmartens/BasisVAE.

The files decoder.py, encoder.py, helpers.py and VAE.py contain the core functionality of the VAE, DeVAE, BasisVAE and BasisDeVAE frameworks.

The file main.py demonstrates the method by i) executing synth_data_gen.py to generate synthetic data and ii) fitting BasisDeVAE to this data as done in the paper.

Core dependencies (excluding PyTorch GPU, which should be configured separately ensuring compatible CUDA support and device drivers) are contained within requirements.txt and can be installed via pip install -r requirements.txt.