Development process¶
Development happens against the master
branch following the
GitHub flow model.
Contributors should use their own forks of the repository. In their fork, they
create feature branches off of master
, and their pull requests should
target the master
branch. Maintainers are responsible for prompt
review of pull requests.
Pull requests against master
trigger automated tests that are run
through Azure DevOps, GitHub Actions, and CircleCI. Additional test suites are
run periodically. When adding new code paths or features, tests are a
requirement to complete a pull request. They should be added in the
test
directory.
Documentation should be provided with pull requests that add or change functionality. This includes comments in the code itself, docstrings, and user guides. For exceptions to this rule the pull request author should coordinate with a maintainer. For changes that fix bugs, add new features, change APIs, etc., i.e., for changes that are relevant to developers and/or users please also add an entry in CHANGES.md in the section corresponding to the next release, since that’s where your change will be included. If you’re a new contributor please also add yourself to AUTHORS.md.
Docstrings should follow numpydoc format. This is a recent decision by the community. The new policy is to update docstrings that a PR touches, as opposed to changing all the docstrings in one PR.
Developer certificate of origin¶
All contributions require you to sign the developer certificate of origin (DCO). This is a developer’s certification in which you declare that you have the right to, and actually do, grant us the rights to use your contribution. We use the exact same one created and used by the Linux kernel developers. You can read it at https://developercertificate.org.
You sign the DCO by signing off every commit comment with your name and email address: Signed-off-by: Your Name <your.email@example.com>
When you submit a pull request, a DCO-bot will automatically determine whether you need to provide a DCO and indicate how you can decorate the PR appropriately (e.g., label, comment).
Manually¶
You can manually sign-off by adding a separate paragraph to your commit message:
git commit -m “Your message
Signed-off-by: Your Name <your.email@example.com>
or
git commit -m “Your message" -m “Signed-off-by: Your Name <your.email@example.com>”
If this feels like a lot of typing, you can configure your name and e-mail in git to sign-off:
git config --global user.name “Your Name”
git config --global user.email “your.email@example.com”
Now, you can sign off using -s
or --signoff
:
git commit -s -m "Your message"
If you find -s
too much typing as well, you can also add an alias:
git config --global alias.c "commit --signoff"
Which allows you to commit including a signoff as git c -m "Your
Message"
.
These instructions were adapted from this blog post.
Automatically¶
You can also fully automate signing off using git hooks, by following the instructions of this stack overflow post.
Advanced installation instructions¶
While working on Fairlearn itself you may want to install it in editable mode. This allows you to test the changed functionality. First, clone the repository locally via
git clone git@github.com:fairlearn/fairlearn.git
To install in editable mode using pip
run
pip install -e .
from the repository root path.
To verify that the code works as expected run
python ./scripts/install_requirements.py --pinned False
python -m pytest -s ./test/unit
Fairlearn currently includes plotting functionality that requires the
matplotlib
package to be installed. Since this is for a niche use case
Fairlearn comes without matplotlib
by default. To install Fairlearn
with its full feature set simply append customplots
to the install
command
pip install -e .[customplots]
Note that the Fairlearn dashboard is built using nodejs and requires additional steps. To build the Fairlearn dashboard after making changes to it, install Yarn, and then run the widget build script.
The Requirements Files¶
The prerequisites for Fairlearn are split between three separate files:
requirements.txt contains the prerequisites for the core Fairlearn package
requirements-customplots.txt contains additional prerequisites for the
[customplots]
extension for Fairlearnrequirements-dev.txt contains the prerequisites for Fairlearn development (such as flake8 and pytest)
The requirements.txt
and
requirements-customplots.txt
files are consumed
by setup.py to specify the dependencies to be
documented in the wheel files.
To help simplify installation of the prerequisites, we have the
install_requirements.py
script which runs pip install
on all three of the above files.
This script will also optionally pin the requirements to any lower bound specified (by changing any
occurrences of >=
to ==
in each file).
Onboarding guide for users of version 0.2 or earlier
Up to version 0.2, Fairlearn contained only the exponentiated gradient method.
The Fairlearn repository now has a more comprehensive scope and aims to
incorporate other methods. The same exponentiated gradient technique is now
the class fairlearn.reductions.ExponentiatedGradient
. While in the past
exponentiated gradient was invoked via
import numpy as np
from fairlearn.classred import expgrad
from fairlearn.moments import DP
estimator = LogisticRegression() # or any other estimator
exponentiated_gradient_result = expgrad(X, sensitive_features, y, estimator, constraints=DP())
positive_probabilities = exponentiated_gradient_result.best_classifier(X)
randomized_predictions = (positive_probabilities >= np.random.rand(len(positive_probabilities))) * 1
the equivalent operation is now
from fairlearn.reductions import ExponentiatedGradient, DemographicParity
estimator = LogisticRegression() # or any other estimator
exponentiated_gradient = ExponentiatedGradient(estimator, constraints=DemographicParity())
exponentiated_gradient.fit(X, y, sensitive_features=sensitive_features)
randomized_predictions = exponentiated_gradient.predict(X)
Please open a new issue if you encounter any problems.
Investigating automated test failures¶
For every pull request to master
with automated tests, you can check
the logs of the tests to find the root cause of failures. Our tests currently
run through Azure Pipelines with steps for setup, testing, and teardown. The
Checks
tab of a pull request contains a link to the
Azure Pipelines page),
where you can review the logs by clicking on a specific step in the automated
test sequence. If you encounter problems with this workflow, please reach out
through GitHub issues.
To run the same tests locally, find the corresponding pipeline definition (a
yml
file) in the devops
directory. It either directly contains
the command to execute the tests (usually starting with
python -m pytest
) or it refers to a template file with the command.
Creating new releases¶
First add a description of the changes introduced in the package version you want to release to CHANGES.md.
It is also best to verify that the Fairlearn dashboard loads correctly. This is slightly involved:
Install the
wheel
package by runningpip install wheel
Create a wheel by running
python setup.py sdist bdist_wheel
from the repository root. This will create adist
directory which contains a.whl
file.Create a new conda environment for the test
In this new environment, install this wheel by running
pip install dist/<FILENAME>.whl
Install any pip packages required for the notebooks using
python ./scripts/install_requirements.py --pinned false
Check that the dashboard loads in the notebooks
We have an Azure DevOps Pipeline which takes care of building wheels and pushing to PyPI. Validations are also performed prior to any deployments, and also following the uploads to Test-PyPI and PyPI. To use it:
Ensure that fairlearn/__init__.py has the correct version set.
Put down a tag corresponding to this version but preprended with
v
. For example, version0.5.0
should be tagged withv0.5.0
.
At queue time, select Test or Production PyPI as appropriate.
As part of the release process, the build_wheels.py
script uses
process_readme.py
to turn all the relative links in the ReadMe file
into absolute ones (this is the reason why the applied tag has be of the form
v[__version__]
). The process_readme.py
script is slightly
fragile with respect to the contents of the ReadMe, so after significant
changes its output should be verified.