ENH: add from_dummies #31795
ENH: add from_dummies #31795
Conversation
pandas/core/reshape/reshape.py
Outdated
| Separator between original column name and dummy variable | ||
| dtype : dtype, default 'category' | ||
| Data dtype for new columns - only a single data type is allowed | ||
| fill_first : str, list, or dict, default None |
There was a problem hiding this comment.
I think I'm -1 on this feature; not sure I see it being super useful outside of niche applications
There was a problem hiding this comment.
@WillAyd thanks for your feedback - when you say "this feature", do you mean from_dummies, or just fill_first?
pandas/core/reshape/reshape.py
Outdated
| Parameters | ||
| ---------- | ||
| data : DataFrame | ||
| columns : list-like, default None |
There was a problem hiding this comment.
I would also rather not provide columns as a keyword; duplicative of the first argument
There was a problem hiding this comment.
Or is this supposed to be the inverse of the prefix argument in get_dummies? If so, I think should just reuse that name
There was a problem hiding this comment.
Sure, I'll change it to prefix
MarcoGorelli
force-pushed
the
from-dummies
branch
from
efbe53f to
303ac44
Compare
|
Hello @MarcoGorelli! Thanks for updating this PR. We checked the lines you've touched for PEP 8 issues, and found: There are currently no PEP 8 issues detected in this Pull Request. Cheers! 🍻 Comment last updated at 2020-04-18 10:00:29 UTC |
MarcoGorelli
force-pushed
the
from-dummies
branch
from
45b5508 to
4be43af
Compare
|
@WillAyd have made some updates. See screenshots in original post for examples. Have removed |
|
@WillAyd thanks for your review, have updated |
|
@MarcoGorelli did you see this impl: #8745 (comment). I also don't see why we would ever return anyting but a Series / Categorical here. can you elaborate? |
I hadn't, thanks for pointing me towards it! @clbarnes that looks good, are you interested in taking over this issue with your implementation? |
|
I can do if you'd prefer that strategy - will need to have a think about the requirements, e.g. what kind of incorrect inputs should it try to gracefully handle, handling masked arrays and NA and so on. That's probably discussion for the issue though. I personally err on the side of being less permissive than pandas is generally so it may end up being pretty constrained, if that's OK with the maintainers. |
Sure, go ahead! I'll close this PR then, though please let me know if for whatever reason you chose not to work on this |
Just a first draft
Will close #8745
Screenshot of examples