API: CategoricalDtype str, repr

[TomAugspurger]

I think that str(cat.dtype) should be changed back to always being 'category'. I've seen several places where people use str(thing.dtype) == 'category' as a way to check for Categoricals. Even subtle things like this arrow PR would break.

So instead of

In [6]: str(pd.Categorical([1, 2, 3]).dtype)
Out[6]: 'CategoricalDtype(categories=[1, 2, 3], ordered=False)'

it would be

Out[6]: 'category'

We can leave __repr__ to be unambiguous.

[jreback]

hmm, I only think you do do that for a CategoricalDtype(None, ordered=False).

[TomAugspurger]

Any reason you would do it just for when categories is None categories?

To be clear, in most places like the output in the console, in the Series / DataFrame repr, you'll still have the informative CategoricalDtype(categories=...) repr. It's only when you call str(x.dtype) that you get 'category'.

[jorisvandenbossche]

Possible option is also to only change str back to 'category', and keeping repr as it is now. That gives the more informative repr in the console, but doesn't break code that used str(dtype).
But, that of course hides a bit that you can do str(dtype) == 'category' (but that is maybe not a bad thing? as we want them to use pd.api.types.is_categorical ?)

[jorisvandenbossche]

Ah, I see that is what you have done in the PR (and stated in the top post)! :-) Should have looked there first and read better.