We should add downwards compatibility testing to ensure scikit-base runs with older versions of numpy and pandas.
While the package itself does not depend on external dependencies, the deep_equals utility does. Hence, the tests for the older versions should be restricted to tests for deep_equals only.
In order to mimick older versions, I would use the uv inbuilt functionality to select a cutoff timepoint.
I would recommend to test at a few months before the numpy 2 release, and also at two time points that cover pandas 1 and pandas 2.
The current tests already cover numpy 2 and pandas 3.
We should add downwards compatibility testing to ensure
scikit-baseruns with older versions ofnumpyandpandas.While the package itself does not depend on external dependencies, the
deep_equalsutility does. Hence, the tests for the older versions should be restricted to tests fordeep_equalsonly.In order to mimick older versions, I would use the
uvinbuilt functionality to select a cutoff timepoint.I would recommend to test at a few months before the
numpy 2release, and also at two time points that coverpandas 1andpandas 2.The current tests already cover
numpy 2andpandas 3.