PolynomialFeatures' docstring does not mention that sparse data is allowed for fit

[mfeurer]

Describe the issue linked to the documentation

The docs for PolynomialFeatures.fit() does not mention that sparse data is allowed. The same holds true for fit_transform.

Suggest a potential alternative/fix

The docs for PolynomialFeatures.fit() and PolynomialFeatures.fit_transform() mention that sparse data is allowed.

[rth]

PolynomialFeatures should work with sparse input

scikit-learn/sklearn/preprocessing/_data.py

Lines 1529 to 1531 in 95d4f08

	If the degree is 2 or 3, the method described in "Leveraging
	Sparsity to Speed Up Polynomial Feature Expansions of CSR Matrices
	Using K-Simplex Numbers" by Andrew Nystrom and John Hughes is

what error are you getting?

[NicolasHug]

I think it's just a doc issue,

instead of

X : array-like, shape (n_samples, n_features)

this should be

X : {array-like, sparse matrix}, shape (n_samples, n_features)

for fit and fit_transform

[mfeurer]

what error are you getting?

There's no error, the docs for fit() and fit_transform() simply don't mention that sparse data is allowed. I expected the docs to say X : array-like or sparse matrix of shape (n_samples, n_features) instead of X : array-like of shape (n_samples, n_features) as for example done by TruncatedSVD

[mfeurer]

This appears to be the case for some other such as (I didn't check all sklearn classes):

• https://scikit-learn.org/stable/modules/generated/sklearn.feature_selection.SelectPercentile.html#sklearn.feature_selection.SelectPercentile.fit
• https://scikit-learn.org/stable/modules/generated/sklearn.feature_selection.SelectKBest.html#sklearn.feature_selection.SelectKBest.fit
• https://scikit-learn.org/stable/modules/generated/sklearn.feature_selection.SelectFpr.html#sklearn.feature_selection.SelectFpr.fit
• https://scikit-learn.org/stable/modules/generated/sklearn.feature_selection.SelectFdr.html#sklearn.feature_selection.SelectFdr.fit

[amueller]

there should probably be a tag and we should test consistency of tag and documentation and actual behavior, right?

[genziano]

Hi all, may I take care of this one?
It would be my first contribution here, and the tag "good first issue" seems appropriate :)

[NicolasHug]

Go ahead @Alemaudit , thanks!

(focus on the docstring, you can ignore the tag comment which is more advanced stuff).

In order to claim the issue for yourself, please comment exactly take below for the CI to automatically assign the issue to you.

[genziano]

take

[genziano]

Thank you @NicolasHug! Done

[genziano]

Quick question:

The .transform() method's doc mentions only the possibility of CSC/CSR sparse matrices. However, it accepts any type of matrices from scipy.sparse, converting the other types to CSR afterward.

scikit-learn/sklearn/preprocessing/_data.py

Lines 1523 to 1524 in f1acf83

	X : array-like or CSR/CSC sparse matrix, shape [n_samples, n_features]
	The data to transform, row by row.

scikit-learn/sklearn/preprocessing/_data.py

Lines 1547 to 1548 in f1acf83

	X = check_array(X, order='F', dtype=FLOAT_DTYPES,
	accept_sparse=('csr', 'csc'))

Is it an idea do document .fit() and transform() the same way?

X : {array-like, sparse matrix}, shape (n_samples, n_features)

[jnothman]

Is it an idea do document .fit() and transform() the same way?

Go ahead