Add arithmetic to categoricals?

[shoyer]

I'm not entirely sure why (perhaps just the trouble of implementing it), but categoricals do not currently support arithmetic:

In contrast to statistical categorical variables, categorical data might have an order (e.g. ‘strongly agree’ vs ‘agree’ or ‘first observation’ vs. ‘second observation’), but numerical operations (additions, divisions, ...) are not possible.

Similarly to string operations (#8627), arithmetic with scalars could very efficient transform categoricals into new categoricals.

In contrast, array + array operations should probably just return a normal array.

CC @JanSchultz @jreback

[jreback]

i believe the thinking was that Categoricals were primarily for strings. But for a numeric (sub-dtype) then certainly this is possible.

can you update with an example of the usecase pls.

[shoyer]

Yeah, I'm actually not entirely sure this is a good idea, given that numbers can already be represented as efficiently in essentially the same space as the categorical codes.

[shoyer]

Actually going to close this, I was getting ahead of myself here. I'm struggling to think of an actual use case for this.

[jreback]

hmm,maybe for a limited set(e.g. < 128) it would save some memory

[jankatins]

Categoricals represents "categories", so in "most" cases nothing with a defined difference (needed for '-' and '+') and no zero (so no '*' or '/'). Even if we have integers as a representation for the categories...

If we really want to have some memory efficient whatever, we can, but then please make that a different class.

[jankatins]

@shoyer Just FYI: Schulz without 'tz', I just found this issues :-)