[
https://issues.apache.org/jira/browse/SPARK-34544?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17292977#comment-17292977
]
Maciej Szymkiewicz commented on SPARK-34544:
--------------------------------------------
The problem we're dealing with here is, that Pandas is not PEP 561 (there is
longer discussion about this issue [on Pandas issue
tracker|https://github.com/pandas-dev/pandas/issues/28142]).
{{DataFrameLike}} is basically a Band-Aid ‒ it allows us to type check things
without type ignoring the pandas package (which in turn would treat Pandas
object as wildcard / {{Any}}, which caused some nasty problems in the past). It
was included on provisional basis until Pandas officially exposes their
annotations and it is more a dev utility than user a facing feature.
As far as I am aware removing it doesn't resolve any of the problems described
here and it makes maintaining annotations harder. For project development needs
we could probably use
[microsoft/python-type-stubs|https://github.com/microsoft/python-type-stubs]
(it wasn't public one I came up with protocol approach), but it is another (and
non-trivial) dependency for the end users and still doesn't address versioning
issue.
> pyspark toPandas() should return pd.DataFrame
> ---------------------------------------------
>
> Key: SPARK-34544
> URL: https://issues.apache.org/jira/browse/SPARK-34544
> Project: Spark
> Issue Type: Bug
> Components: PySpark
> Affects Versions: 3.1.1
> Reporter: Rafal Wojdyla
> Assignee: Maciej Szymkiewicz
> Priority: Major
>
> Right now {{toPandas()}} returns {{DataFrameLike}}, which is an incomplete
> "view" of pandas {{DataFrame}}. Which leads to cases like mypy reporting that
> certain pandas methods are not present in {{DataFrameLike}}, even tho those
> methods are valid methods on pandas {{DataFrame}}, which is the actual type
> of the object. This requires type ignore comments or asserts.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
