DEPRECATION NOTICE: the scikit-learn wheel-building infrastructure has been moved to the main repo. It now uses the cibuildwheel system instead of the multibuild system.
Repository to build scikit-learn wheels.
Edit azure/windows.yml and azure/posix.yml to change the BUILD_COMMIT environment variable to set the name of the git tag to build, commit and push (to master).
Azure Pipelines should automatically build and test that version on Windows, Linux and OSX for various versions of Python (both 32 bit and 64 bit).
If the tests pass, the resulting wheels should show up on:
- https://anaconda.org/scikit-learn-wheels-staging for release wheels staging;
- https://anaconda.org/scipy-wheels-nightly for nightly builds.
The following tool can be useful to download all the wheels for a specific release:
https://github.com/ogrisel/wheelhouse-uploader
and then use twine to publish all the wheels along with the locally built source tarball of the release all at once to PyPI.