Reviewing the release notes of sklearn (here), I realized that the version that is being checked in CI is 0.20.3 wich has an official support for Python 2.7 and 3.4 only.
This is inconsistent, because metric_learn only supports Python 3.6+.
Then, the minimum version of sklearn allowed should be 0.21.3 wich is the older version officially supporting Python 3.6. Also the older version with documentation (here).
There should also be restrictions to numpy and scipy, to at least match the minimum versions required for sklearn 0.21.3.
For the CI, only Python 3.6 and 3.7 should be checked, as only these verions are officially supported by sklearn 0.21.3
Reviewing the release notes of sklearn (here), I realized that the version that is being checked in CI is 0.20.3 wich has an official support for Python 2.7 and 3.4 only.
This is inconsistent, because metric_learn only supports Python 3.6+.
Then, the minimum version of sklearn allowed should be 0.21.3 wich is the older version officially supporting Python 3.6. Also the older version with documentation (here).
There should also be restrictions to numpy and scipy, to at least match the minimum versions required for sklearn 0.21.3.
For the CI, only Python 3.6 and 3.7 should be checked, as only these verions are officially supported by sklearn 0.21.3
Modified: setup.py, readme, docs and CI