FIX degree in polynomial kernel should be a floating number

[NolantheNerd]

Reference Issues/PRs

None - this fix is not associated with an issue.

What does this implement/fix? Explain your changes.

This fix only changes the docs.

The SpectralClustering class' degree parameter is constrained to positive integers, but the SpectralClustering class' docstring specifies that degree can be a float.

I've updated the docstring to require that degree be passed as an int.

[github-actions]

✔️ Linting Passed

All linting checks passed. Your pull request is in excellent shape! ☀️

_{Generated for commit: 588fd5e. Link to the linter CI: here}

[glemaitre]

I think the documentation is correct but the code is wrong.

The degree will be used by a the polynomial kernel:

scikit-learn/sklearn/metrics/pairwise.py

Lines 1355 to 1369 in 93fa00c

	@validate_params(
	{
	"X": ["array-like", "sparse matrix"],
	"Y": ["array-like", "sparse matrix", None],
	"degree": [Interval(Real, 1, None, closed="left")],
	"gamma": [
	Interval(Real, 0, None, closed="left"),
	None,
	Hidden(np.ndarray),
	],
	"coef0": [Interval(Real, None, None, closed="neither")],
	},
	prefer_skip_nested_validation=True,
	)
	def polynomial_kernel(X, Y=None, degree=3, gamma=None, coef0=1):

and this is constrained to be a float.

@jeremiedbb I see that I missed this thing in the following review: #24035

We also tolerate an integer at 0 that will be excluded by the kernel itself. I assume that we could fore all the degrees parameter to "degree": [Interval(Real, 0, None, closed="left")]

@jeremiedbb WDYT?

[glemaitre]

Interacting IRL with @jeremiedbb, we agreed that this is a regression and that we should use a Real number instead of Integral her. @NolantheNerd Do you want to contribute this change?

[NolantheNerd]

Done! Ready for review!

[glemaitre]

Can you also update:

• the degree parameter from KernelPCA to be a real as well
• the degree parameter from KerelRidge to be a real as well

In addition, we will need an entry in the changelog to acknowledge the regression.

[NolantheNerd]

Changes complete! Now for cluster.SpectralClustering,kernel_ridge.KernelRidge and decomposition.KernelPCA:

• The docstring reflects that degree is a float.
• The default value has been changed from 3 -> 3.0 in the docstring to further clarify that floats are permitted.
• The default value in the constructor has been changed from 3 -> 3.0.

There doesn't seem to be a consistent pattern for integer/float disambiguation in the codebase, for example, in the KernelPCA docstring, both 1 and 1.0 are used for float values. @glemaitre Is there a "right" way to do it?

scikit-learn/sklearn/decomposition/_kernel_pca.py

Lines 62 to 73 in 4ca0196

	coef0 : float, default=1
	Independent term in poly and sigmoid kernels.
	Ignored by other kernels.
	kernel_params : dict, default=None
	Parameters (keyword arguments) and
	values for kernel passed as callable object.
	Ignored by other kernels.
	alpha : float, default=1.0
	Hyperparameter of the ridge regression that learns the
	inverse transform (when fit_inverse_transform=True).

[glemaitre]

There doesn't seem to be a consistent pattern for integer/float disambiguation in the codebase

In this case both integral an float have the same behaviour so there is no need for disambiguation.

Sometimes we have float and integral that does not behave the same but usually the range of the accepted values will be different and the parameter validation will be here to help.

[glemaitre]

I only think that we can keep 3 instead of 3.0 since this is already a valid Real.

Otherwise LGTM. Thanks @NolantheNerd

[OmarManzoor]

LGTM. Thanks @NolantheNerd

[OmarManzoor]

@NolantheNerd I don't see any conflicts but Github is complaining that there are. Could you kindly check if there are any conflicts with main or otherwise maybe push an empty commit to retrigger the CI?

[NolantheNerd]

@OmarManzoor looks like there was a conflict in the news file. I've fixed the conflict.

[glemaitre]

Merging then. Thanks @NolantheNerd