REF: Rename mode.nullable_backend to mode.dtype_backend #50291
Conversation
mroeschke
added
Refactor
There was a problem hiding this comment.
Thanks @mroeschke!
FWIW mode.dtype_backend makes sense to me and seems like an improvement over mode.nullable_backend. It might also be worth considering mode.dtypes, which would be nice because of it's conciseness, though maybe it's too concise. I'm curious to hear what you and others think.
|
@phofl any thoughts on |
|
The nice thing about the old name: It implies that you have to use nullable dtypes that the option can take effect. If you don't use nullable_dtypes then nullable_backend does not make much sense. The new name implies that it can work globally, which is not possible right now. In general, I'd be ok with this, but we have to clarify in the docs that this works only in specific situations right now |
|
Agreed with the concern, at least for 2.0 where we will be paring this closely with nullable types, but I like that this change future-proofs this option to be more applicable to the non-nullable types in the future |
pandas/core/config_init.py
Outdated
| @@ -539,7 +539,7 @@ def use_inf_as_na_cb(key) -> None: | |||
| The default storage for StringDtype. | |||
| """ | |||
|
|
|||
| nullable_backend_doc = """ | |||
| dtype_backend_doc = """ | |||
| : string | |||
| The nullable dtype implementation to return. | |||
There was a problem hiding this comment.
I am ok with this too, but I'd like to make this more prominent here that this only works with use_nullable_dtypes as of right now
There was a problem hiding this comment.
I think it's also used with convert_dtypes() now and will (hopefully) be used in more places in the future. Instead of explicitly listing where this option is used, maybe we have some more generic statement about it still being experimental and only used in certain methods -- see the API docs for a specific method to see if it supports this option?
There was a problem hiding this comment.
Yep this sounds good to me. I am happy as long as we clarify that it is not generic yet.
There was a problem hiding this comment.
Sounds good. Clarified here that it's only applicable where documented
|
thx @mroeschke |
From the dev call yesterday, it was discussed to make the new global configuration for 2.0
mode.nullable_backend="pyarrow"|"pandas"more generically named. I have changed it tomode.dtype_backendas we may decide to not always associate this option with theuse_nullable_dtypekeyword argument in the IO readerscc @phofl @jrbourbeau