Mention possibility of regression targets in warning about unique classes >50% of n_samples #31689
Conversation
sklearn/utils/multiclass.py
Outdated
| "of samples. Note this may be because the data belongs to a regression " | ||
| "problem, not a classification problem.", |
There was a problem hiding this comment.
That would only be triggered by a regression target containing only integers and with an integer dtype. Is it worth it ?
There was a problem hiding this comment.
Well technically integer float values are "multiclass", but I get your point...
type_of_target([1.0, 0.0, 3.0])
'multiclass'There was a problem hiding this comment.
Maybe it is too rare. Happy to close
There was a problem hiding this comment.
Right, I misread the previous code block, integer float values end-up as multiclass.
The original motivation for the warning was not about confusion with regression problems according to #16399 (comment).
However this comment #26335 (comment) goes in your direction. So I think your addition is worth.
There was a problem hiding this comment.
Hmm yes. Christian also mentions it later: #26335 (comment)
sklearn/utils/multiclass.py
Outdated
| "of samples. Note this may be because the data belongs to a regression " | ||
| "problem, not a classification problem.", |
There was a problem hiding this comment.
Small nit:
| "of samples. Note this may be because the data belongs to a regression " | |
| "problem, not a classification problem.", | |
| "of samples. The target data could represent a regression " | |
| "problem, not a classification problem.", |
There was a problem hiding this comment.
This is nicer thank you. What do you think about:
"y could represent a regression problem, not a classification problem." ..?
(I am happy either way)
There was a problem hiding this comment.
Your suggestion works for me to.
…sses >50% of n_samples (scikit-learn#31689) Co-authored-by: Jérémie du Boisberranger <jeremie@probabl.ai>
…sses >50% of n_samples (scikit-learn#31689) Co-authored-by: Jérémie du Boisberranger <jeremie@probabl.ai>
…sses >50% of n_samples (scikit-learn#31689) Co-authored-by: Jérémie du Boisberranger <jeremie@probabl.ai>
…sses >50% of n_samples (scikit-learn#31689) Co-authored-by: Jérémie du Boisberranger <jeremie@probabl.ai>
…sses >50% of n_samples (scikit-learn#31689) Co-authored-by: Jérémie du Boisberranger <jeremie@probabl.ai>
…sses >50% of n_samples (scikit-learn#31689) Co-authored-by: Jérémie du Boisberranger <jeremie@probabl.ai>
Reference Issues/PRs
Addresses #26335 (comment)
What does this implement/fix? Explain your changes.
In #26335 we added a warning when unique classes >50% of
n_samples. This adds a note that it could also be due to the data being from a regression problem.Any other comments?