I'd be in favor of whatever the conclusion of the same question on the scipy-dev thread be ( https://mail.python.org/pipermail/scipy-dev/2020-August/024318.html) which seems to be dropping 3.6 for the next release. But I don't think we're going to be supporting only 3 latest releases especially since Python has moved to the annual release cycle. On Fri, Aug 21, 2020 at 4:23 PM Martin Thoma <info@martin-thoma.de> wrote:

Hi :-)

Did you discuss at some point a policy which Python versions you want to support? I see that scikit-learn supports at the moment 3.6, 3.7, and 3.8 (badge in README). In a couple of weeks (October?) there will be 3.9, but I don't see any issue opened discussing 3.9

A minor python increment (e.g. 3.8 -> 3.9) will now happen every year, IIRC. So maybe it would be nice to just say "we support the latest 3 Python versions". In the very least, this would mean that wheels are published and that the CI pipeline is run for the newer versions.

I think it would also be nice to "deprecate" old features and drop support explicitly. For example, if scikit-learn dropped support for 3.6 we could use future annotations. The implication of such a drop of support would be a major version increase.

Best regards, Martin _______________________________________________ scikit-learn mailing list scikit-learn@python.org https://mail.python.org/mailman/listinfo/scikit-learn