PrefSampling

Overview

PrefSampling is a lightweight Python library that provides preference samplers. These are algorithms that generate random preferences based on precisely defined statistical cultures. We consider different type of preferences:

• Ordinal: preferences are expressed as rankings of the candidates;
• Approval: preferences are expressed by indicating a set of approved candidates.

This package is part of the Guide to Numerical Experiments on Elections in Computational Social Choice.

Installation

The package can be installed from PyPI using:

pip3 install prefsampling

Documentation

The complete documentation is available here.

Citing our Work

If you are using this package we kindly ask you to cite the following reference to credit our work link.

Boehmer N., Faliszewski P., Janeczko Ł., Kaczmarczyk A., Lisowski G., Pierczyński G., Rey S., Stolicki D., Szufa S., Wąs T. (2024).
Guide to Numerical Experiments on Elections in Computational Social Choice.
arXiv preprint arXiv:2402.11765.

Development

Setting up the development mode

We are more than happy to receive help with the development of the package. If you want to contribute, here are some elements to take into account.

First, install the development dependencies by running the following command:

pip install -e ".[dev]"

Conventions

We try to enforce uniformity within the package. Here are some general guidelines.

• All samplers have num_voters and num_candidates as their two first positional arguments
• All samplers accept a seed parameter to set the seed of the random number generator

Within the package, the samplers are organised in modules based on the ballot format they generate. The prefsampling.core module is used for features used across samplers. Within the submodule corresponding to the ballot format, there is a Python file for each family of samplers. All the samplers are imported and appear in the __all__ variable of the __init__.py file of the corresponding module (defined by the ballot format).

Tests

The tests are run with unittest. Simply run the following command to launch the tests:

python -m unittest

The structure of the test module follows that of the package. There is one submodule per ballot format we sample. Within the submodule, there is one file per statistical culture.

At the submodule level, there is a file test_all_ballotformat_samplers.py that gathers the test that are common to all samplers of the given ballot format.

In the file corresponding the statistical culture, there is a function that returns all the samplers (with their arguments set) that are used as test cases, together with the tests that are specific to the sampler.

When a new sampler is added to the package, it needs to be added in several places within the test module:

• A file test/ballotformat/test_ballotformat_culturename.py defining the tests specific to the sampler and the functions to use for the tests (called random_ballotformat_culturename_samplers).
• In test_all_ballotformat_samplers.py, add the functions for the sampler to the random_ballotformat_samplers() function.
• If it is a sampler for actual ballots (i.e., not points in space or trees), add the functions for the samplers to the random_samplers() in the file test/test_all_samplers.py.

Validation

We aim at statistically validating the samplers we provide. All the code necessary to run the validation is gathered in the validation folder of the repository.

When a new sampler is added to the package, proceed as follows:

• Create the corresponding file in the validation/ballotformat/ folder.
• In this file, define a class that inherit from the validation.validator.Validator. This requires you to define a set of methods used to compute the theoretical probabilities of the outcomes of the samplers.
• Add the validator in the run.py file.
• Run the run.py file (you may want to comment out some parts).
• Copy the generated graphs in the correct place of the doc-source/source/validation_plots folder.
• Update the doc-source/source/validation.rst file accordingly.

Documentation

The doc is generated using sphinx. We use the numpy style guide. The napoleon extension for Sphinx is used and the HTML style is defined by the Book Sphinx Theme.

To generate the doc, first move inside the docs-source folder and run the following:

make clean 
make html

This will generate the documentation locally (in the folder docs-source/build). If you want the documentation to also be updated when pushing, run:

make githubclean
make github

After having pushed, the documentation will automatically be updated. Note that the github directive may not work on Windows.

Publishing on PyPI

The pipeline between GitHub and PyPI is automatised. To push a new version do the following:

• Update the pyproject.toml with the new version number.
• Update the prefsampling/__init__.py with the new version number.
• On GitHub, create a new release tagged with the new version number (only admins can do that).
• You're done, the new version of the package is automatically pushed to PyPI after the creation of a GitHub release.