ENH Warn on irrelevant parameters for linear kernel in SVC #30174
Conversation
There was a problem hiding this comment.
This also would need tests, a starting point was provided in #19630
… into svm-warning
| @@ -217,7 +217,25 @@ def fit(self, X, y, sample_weight=None): | |||
| "X and y have incompatible shapes.\n" | |||
| + "X has %s samples, but y has %s." % (n_samples, y.shape[0]) | |||
| ) | |||
|
|
|||
| if self.kernel == "linear": | |||
| if self.gamma != "scale": | |||
There was a problem hiding this comment.
I think it's odd to check for the default value here, since the user might be setting these default values explicitly. One alternative is to have the default value as None, and here check if it's None.
Then it begs the question, do we automatically assign a default value to gamma when it's None or not. As a consequence, the question becomes, should this be valid code?
SVC(kernel="rbf")Or should the user always give a gamma as:
SVC(kernel="rbf", gamma=.1)There was a problem hiding this comment.
@adrinjalali I think it's better to require a value of gamma to be clearly specified. Since the default value is already None, we might not want to have another default value when None is chosen. For me it's strange to have two default values for a single function call.
Reference Issues/PRs
closes #28641
closes #19630
Fixes #19614
What does this implement/fix? Explain your changes.
Add user warnings when irrelevant parameters are set for linear kernel
Any other comments?