pawel-czyz / labelshift

Bayesian Quantification with Black-Box Estimators

Bayesian quantification with black-box estimators

Quantification is the problem of estimating the label prevalence from an unlabeled data set. In this repository we provide the code associated with our manuscript, which can be used to reproduce the experiments.

Installation

We recommend using Micromamba to set a new Python 3.11 environment. Then, the package can be installed with:

$ pip install -e .

To reproduce the experiments, install Snakemake using the instructions provided. Then, install additional dependencies:

$ pip install -r requirements.txt

The experiments can be reproduced by running:

$ snakemake -c4 -s workflows/WORKFLOW_NAME.smk

Citation

In order to cite this code, please use:

@article{
    Ziegler-Czyz-2024-Bayesian-Quantification,
    title={Bayesian Quantification with Black-Box Estimators},
    author={Albert Ziegler and Pawe{\l} Czy{\.z}},
    journal={Transactions on Machine Learning Research},
    issn={2835-8856},
    year={2024},
    url={https://openreview.net/forum?id=Ft4kHrOawZ}
}