DEPR: Make `dtype="category"` always imply ordered=False

[mroeschke]

Currently one can specify dtype="category" to specify data as CategoricalDtype, and since it does not convey ordered information, we treat it as ordered=None internally in order to no-op data that is already categorical. Therefore internally, there are scenarios where we need to be mindful if we have ordered=None vs ordered=True/False

To simplify this gymnastics, it would be great if dtype="category" always meant CategoricalDtype(ordered=False). I think we would only need to issue a deprecation warning when specifying dtype="category" when the data is categorical and ordered=True

[ZanirP]

Hi! I would love to take this!

Just to clarify, the goal is to make dtype="category" always mean CategoricalDtype(ordered=False), with a deprecation warning when applied to CategoricalDtype(ordered=True).

So for example:

cat_ordered = pd.Categorical(["low", "medium", "high"], ordered=True)
categorical_data = cat_ordered.astype("category")

Current behavior is: CategoricalDtype(categories=['high', 'low', 'medium'], ordered=True, categories_dtype=object)
Expected behavior is: CategoricalDtype(categories=['high', 'low', 'medium'], ordered=False, categories_dtype=object) with a deprecation warning being raised (to inform users that ordering is being removed)?

Let me know if this sounds good!

[ZanirP]

take