The VampPrior Mixture Model

This code generates all results in our manuscript:

@misc{stirn2024vampprior,
      title={The VampPrior Mixture Model}, 
      author={Andrew Stirn and David A. Knowles},
      year={2024},
      eprint={2402.04412},
      archivePrefix={arXiv},
      primaryClass={cs.LG}
}

Package Requirements

Our Dockerfile creates the computational environment from TensorFlow's official image for version 2.15.0 with GPU support and uses requirements.txt to install additionally required packages.

Reproducing Experiments

Run clustering_experiments_run.sh to produce Table 1 and Figures 1 and 3.

Run elbo_surgery_run.sh to produce Table 2.

Run single_cell_experiments_run.sh to produce Tables 3 and 4 and Figures 2, 4-6.

Repository Overview

priors.py implements the our Bayesian GMM and VMM.

models.py contains the VAE models.

single_cell_models.py has our implementation of scVI (Lopez et al., 2018).

clustering_*.py are used by clustering_experiments_run.sh.

elbo_surgery_*.py are used by elbo_surgery_run.sh.

single_cell_*.py are used by single_cell_experiments_run.sh.