Pandas 3.0 may impact the automatic text treatment in sklearn

[folmos-at-orange]

Description

Pandas 3.0 will have PyArrow strings as default instead of object. Does this impact the way we currently have to specify Text columns in the sklearn estimators ?

[popescu-v]

IMHO it does not, because the current heuristic also checks that len(feature) > 100. However, it might be a safe strategy to document that (starting from Pandas 3.0) string features whose length is greater than 100 are automatically typed as "Text" rather than as "Categorical" and that, if the latter typing is required, then the feature in question should be recoded.