Fix column_transformer to use fitparams like Pipeline #21311
Conversation
| fit_params = {'standard_scaler__sample_weight': df['equal_sample_weight']} | ||
| ct_wEqualWeight = ct.fit_transform(X=df[['x1']], y=df['y'], **fit_params) | ||
|
|
||
| assert_array_equal(sc1_xWeight, ct_xWeight, err_msg= "These should be equal") |
There was a problem hiding this comment.
pep8: there shouldn't be a space after "=" here. But you can also just remove the err_msg I think.
| return "(%d of %d) Processing %s" % (idx, total, name) | ||
|
|
||
| def _fit_transform(self, X, y, func, fitted=False, column_as_strings=False): | ||
| def _check_fit_params(self, **fit_params): |
There was a problem hiding this comment.
Is this copied from pipeline or somewhere else?
There was a problem hiding this comment.
yes, it came from pipeline. I tried to make it work in the same way.
|
Thanks for the PR, looks good overall, I think :) This makes sense to unblock this use-case but really we need to fix how to do this properly, cc @adrinjalali ;) |
|
Thanks for the PR. But I would rather prioritize sample props and #21284 for the next release, in which case we wouldn't need this solution. |
|
@adrinjalali that makes sense. How would the code look like for a simple pipeline with a column transformer to pass sample_weight everywhere? |
|
So taking an example from here: from sklearn.compose import ColumnTransformer
from sklearn.feature_extraction.text import CountVectorizer
from sklearn.preprocessing import OneHotEncoder
column_trans = ColumnTransformer(
[('categories', OneHotEncoder(dtype='int').fit_requests(sample_weight=True), ['city']),
('title_bow', CountVectorizer().fit_requests(sample_weight=True), 'title')],
remainder='drop', verbose_feature_names_out=False)And if you put this column transformer in a pipeline, you don't need to do anything extra, since the consumer has already requested the metadata. Then if you call |
|
Sorry, just getting back to this. That will work, thank you! |
|
ColumnTransformer now supports metadata routing. |
Reference Issues/PRs
Fixes #19465.
What does this implement/fix? Explain your changes.
This fix implements fitparams in the same way as Pipelines to meet parity
Any other comments?