FIX handle column names renaming in ColumnTransformer - #28262
Conversation
|
I could also use the code snippet of the original issue as an integration test. However, they weird stuff to be investigated in between:
So more to follow. |
ogrisel
left a comment
ogrisel
left a comment
There was a problem hiding this comment.
Suggestion about the error message.
Co-authored-by: Olivier Grisel <olivier.grisel@ensta.org>
| continue | ||
| names_out = feature_names_outs[names_idx : names_idx + X.shape[1]] | ||
| adapter.rename_columns(X, names_out) | ||
| names_idx += X.shape[1] |
There was a problem hiding this comment.
I find this code ugly but I don't see how to improve it without reshaping boilerplate
|
I think it is ready for a review @ogrisel @adrinjalali There quite some boilerplate code for generating the error but I think this is OK. For the rest, I'm just annoyed between having a flat list and a list of list. But I don't see how to simplify if I want to avoid touching more code that is not related to this issue. |
ogrisel
left a comment
ogrisel
left a comment
There was a problem hiding this comment.
Some more comments but otherwise LGTM.
| raise ValueError( | ||
| "Concatenating DataFrames from the transformer's output lead to" | ||
| " an inconsistent number of samples. The output may have Pandas" | ||
| " Indexes that do not match." |
There was a problem hiding this comment.
Note: we could actually check if that last sentence hold to make to the error message more precise (and not use "may" when it's not the case).
But let's keep that for a follow-up PR ;)
Co-authored-by: Olivier Grisel <olivier.grisel@ensta.org>
adrinjalali
left a comment
adrinjalali
left a comment
There was a problem hiding this comment.
This seems like there must be a better way to do it 😅
| continue | ||
| names_out = feature_names_outs[names_idx : names_idx + X.shape[1]] | ||
| adapter.rename_columns(X, names_out) | ||
| names_idx += X.shape[1] |
There was a problem hiding this comment.
_num_features and _num_samples instead of .shape?
There was a problem hiding this comment.
Here we play with container that are dataframe-like container. So they will implement the shape.
…8262) Co-authored-by: Olivier Grisel <olivier.grisel@ensta.org>
…8262) Co-authored-by: Olivier Grisel <olivier.grisel@ensta.org>
Co-authored-by: Olivier Grisel <olivier.grisel@ensta.org>
3 participants
closes #28260
The adapters in charge of renaming columns with
set_outputto a DataFrame-like object was failing in case that the original DataFrame-like as duplicated columns (that is possible inpandas.We inverse the stacking and and renaming to avoid potential column duplicate that should not happen.