Explain the change
The PR should substantiate the change: through benchmark of performance or example of usage
Make sure the code is commented and well documented.
Build the doc locally and check its correctness
Add time and space complexity
Add tutorial and relevant material
Take part in maintenance and support
PR Communication
PR Prefix
MAINT and MNT: general project maintenance, e.g. private code edits that doesn’t impact users
FEA: new feature
EHN and ENH: improving an existing feature
FIX: fixture
DOC: documentation
CI: continuous integration
TST: tests
POC: proof of concept
PERF:Â EHNÂ of performance
PR Title
Sometimes <FIX “PR_NAME”> is enough, but <FIX #PR_NUMBER> is never a good title
Add links to related PR, e.g. closing #1234
Testing
Run pytest on changed files
Test a single module: pytest sklearn/tests/test_common.py -v -k LogisticRegression
Display variable on error: pytest -l
Run pdb shell on error: pytest --pdb
Run ipdb shell on error: pytest --pdbcls=IPython.terminal.debugger:TerminalPdb --capture no
Doesn’t capture print: pytest -s
Check unittest coverage (at least 80%)
Shell
Copy
$ pip install pytest pytest-cov
$ pytest --cov sklearn path/to/tests_for_package
Run static typing analysis with mypy sklearn
Using # type: ignore annotation can be a workaround for a few cases
Auto-format code: black .
Make sure the code is flake8 compatible: git diff upstream/main -u -- "*.py" | flake8 --diff