FEA return final cross-validation score in SequentialFeatureSelector #31483
Conversation
doc/whats_new/upcoming_changes/sklearn.feature_selection/31483.feature.rst
Outdated
Show resolved
Hide resolved
|
See latest changes to address your comments |
doc/whats_new/upcoming_changes/sklearn.feature_selection/31483.feature.rst
Outdated
Show resolved
Hide resolved
….feature.rst update based on PR suggestions Co-authored-by: Adrin Jalali <[email protected]>
…_cv_score` to return raw values instead of mean
There was a problem hiding this comment.
I don't mind the implementation as is, but I do wonder its usecases and whether it's useful to enough users.
Tagging for a second opinion: @OmarManzoor @adam2392
| scores : ndarray of shape (n_splits,) | ||
| Array of cross-validation scores for each split. | ||
| """ | ||
| _raise_for_params(params, self, "get_final_cv_score") |
There was a problem hiding this comment.
please have a test for this.
|
When I approved this I considered it being added as an attribute but since that increases the fit time I am not so sure about having a separate function that will still need to be called separately. Wouldn't that kind of be similar to just calling the code within the function? I guess if it adds some convenience to users we can add it. |
Not sure which function you mean. |
|
There was a problem hiding this comment.
IIUC, this is purely a convenience function right?
The computation time to get the answer that you'd want is the same with or without the function.
In that case, my main criterion would be looking at whether this makes the API more usable. Is this function name also present in other feature selectors? If so, let's add it imo. If not, shouldn't we consolidate?
| @@ -193,6 +193,21 @@ def __init__( | |||
| self.cv = cv | |||
| self.n_jobs = n_jobs | |||
|
|
|||
| def _get_cv(self, y): | |||
There was a problem hiding this comment.
I don't see why this function is needed. Perhaps I'm missing something?
There was a problem hiding this comment.
It was a suggestion in one of the comments but basically we had duplicate code in 2 places (cv = check_cv(self.cv, y, classifier=is_classifier(self.estimator))) so it was moved into a function to clean it up
Reference Issues/PRs
What does this implement/fix? Explain your changes.