Code style

Github conventions

Because the Fairlearn team squash merges pull requests, you do not need to put much effort into the commit messages you submit when pushing code.

Titles should be descriptive and include one of the following prefixes: * DOC: any documentation-related PRs (including the user guide, API reference, and other website-related PRs). * MNT: code maintenance (refactoring, improve efficiency, etc.). * CI: anything related to our automated tests (nightly builds, CircleCI, releases, etc.). * FIX: bug fixes. * FEAT/ENH: adds a new feature or removes a feature in the codebase.

Test coverage is checked with codecov and line(s) missing tests will show up in CI as a failure. Therefore, we recommend contributors ensure all new content they introduce has adequate test coverage.

Code conventions

Linting

We recommend using a linter to check your code before you submit a PR. We use ruff to check for PEP8 compatibility issues. You can either follow the guidelines, or you can run black on your code. The generated formatting by black is compatible our with formatting requirements. You can configure your IDE to use black to format your code. Please refer to your IDE’s instructions for further details. Attaining the project compatible linting is also possible without changing your local setup. You can enable the pre-commit hooks, which will run the linters described above with the settings defined in the [pyproject.toml](fairlearn/fairlearn). The installation instructions are described in step 6 of `"Contributing a Pull Request"`<https://fairlearn.org/main/contributor_guide/development_process.html#contributing-a-pull-request>`_ .

Considerations for new methods

If you are introducing new estimators to the Fairlearn, you must ensure the estimator is fully compatible with scikit-learn (defined here ). For more resources on how to develop scikit-learn estimators, review this post by one of the Fairlearn maintainers.

The Fairlearn team is in the process of swapping out the pandas library for narwhals for data manipulation tasks. If you are contributing code that includes Pandas, we recommend you use Narwhals instead to stay ahead of this effort.

Because there are many complex sources of unfairness — some societal and some technical — it is not possible to fully “debias” a system or to guarantee fairness. In Fairlearn, we therefore try to avoid naming mitigation techniques in a way that could suggest they offer a simple fix towards a “fair” model. Instead, we opt for descriptive names, such as “ThresholdOptimizer” (rather than e.g. “FairThresholder”) and “CorrelationRemover” (instead of e.g. “BiasRemover”).

Test coverage

For more information about Fairlean’s default coverage settings check the codecov documentation.