API: Change str for CategoricalDtype to category

[TomAugspurger]

Better compatibility with older versions

Closes #17782

[codecov]

Codecov Report

Merging #17783 into master will decrease coverage by 0.01%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master   #17783      +/-   ##
==========================================
- Coverage   91.24%   91.22%   -0.02%     
==========================================
  Files         163      163              
  Lines       49916    49910       -6     
==========================================
- Hits        45544    45530      -14     
- Misses       4372     4380       +8

Flag	Coverage Δ
#multiple	89.02% <ø> (ø)	⬆️
#single	40.23% <ø> (-0.08%)	⬇️

Impacted Files	Coverage Δ
pandas/core/dtypes/dtypes.py	95.14% <ø> (+0.19%)	⬆️
pandas/io/gbq.py	25% <0%> (-58.34%)	⬇️
pandas/core/frame.py	97.73% <0%> (-0.1%)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 48d0460...c1774b6. Read the comment docs.

[jreback]

I am not sure this is a great idea, what is the purpose here?

[TomAugspurger]

There's, lot's of code out there that expects str(dtype) to be 'category ', and an established tradition in Python of str(obj) being simple, while repr(obj) is more complicated.

Why are you against it?

[jreback]

this loses informtion, no reason why for that. where does code == 'str(dtype)` live? IOW can you show me an example from a project?

[TomAugspurger]

The information is still there... It's only when you call str(dtype). I linked to an example in the issue - https://github.com/dask/fastparquet/pull/200/files - dask/dask@f8e382c - https://github.com/apache/arrow/pull/1161/files#diff-4b312e1fbf7b4c070f9788be9db33b20R198

…

On Thu, Oct 5, 2017 at 7:17 AM, Jeff Reback ***@***.***> wrote: this loses informtion, no reason why for that. where does code == 'str(dtype)` live? IOW can you show me an example from a project? — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#17783 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABQHIuzmNIBNpLZDD9nVR_lYyGuDyUYQks5spMjigaJpZM4PtwTu> .

[jreback]

the if str(dtype) == 'category' was never a way to do this. to be honest I am in favor of breaking this.

[TomAugspurger]

Ok, I'm +1 for this change, and so is Joris.

[jorisvandenbossche]

It maybe never really was meant to do that, but before is_categorical was consistently publicly exposed, I think it was a sensible thing to do.
Personally I don't find undoing the loss of information when converting dtype to string worth it to potentially break people's code.

[jreback]

ok then, merge away.

[jorisvandenbossche]

Another reason that I think this was a good change, is because it is used in the 'dtypes' overview. Before this PR you got:

In [19]: df.dtypes
Out[19]:
cats    CategoricalDtype(categories=['a', 'b', 'c'], o...
vals                                                int64
dtype: object

which I think is not ideal (although that might have been solved in another way as well).

But that did me think, we could think of some kind of 'shorter but still informative' repr like datetimetz has (although once you have more categories, it will always be truncated, which is not that nice in a default repr I think).