Fix nightly with pd.DataFrame.rename calling for MultiIndex.fillna #1449
Conversation
|
/pandas_nightly |
|
https://github.com/pandas-dev/pandas-stubs/actions/runs/18793416801/job/53628688389 Nightly pandas is passing |
| # Apparently all of these calls are accepted by pandas | ||
| check(assert_type(df.rename(columns={None: "b"}), pd.DataFrame), pd.DataFrame) | ||
| check(assert_type(df.rename(columns={"": "b"}), pd.DataFrame), pd.DataFrame) | ||
| check(assert_type(df.rename(columns={(2, 1): "b"}), pd.DataFrame), pd.DataFrame) |
There was a problem hiding this comment.
Could you change the stub files and invalidate this usage?
There was a problem hiding this comment.
That would actually break, the issue is at run time when you try to rename a dataframe with an Index when the mapper is designed for a multiindex. Since we do not have the capacity to differenciate 'dataframe with index' and 'dataframe with multiindex' you can't force that behavior in the stubs. Do you see what I mean?
There was a problem hiding this comment.
One possible solution would be to only allow rename(column=d), where d is a dictionary containing tuples as keys and values of equal length, so that columns={(2, 1): "b"} will be forbidden.
However I couldn't work out how to implement this. For now, I'll leave it as it is.
| # Apparently all of these calls are accepted by pandas | ||
| check(assert_type(df.rename(columns={None: "b"}), pd.DataFrame), pd.DataFrame) | ||
| check(assert_type(df.rename(columns={"": "b"}), pd.DataFrame), pd.DataFrame) | ||
| check(assert_type(df.rename(columns={(2, 1): "b"}), pd.DataFrame), pd.DataFrame) |
There was a problem hiding this comment.
One possible solution would be to only allow rename(column=d), where d is a dictionary containing tuples as keys and values of equal length, so that columns={(2, 1): "b"} will be forbidden.
However I couldn't work out how to implement this. For now, I'll leave it as it is.
2 participants
assert_type()to assert the type of any return valuePR replaces the
DataFrame.renamewhen using a multi index in the replacer while the dataframe only hasIndex.