IsotonicRegression with 2D input and n_features=1

[rth]

IsotonicRegression works only on 1D data, however I think it should not fail when provided a 2D input with n_features=1.

Example

In other words the following example (from the docstring) should work without having to flatten X,

    >>> from sklearn.datasets import make_regression
    >>> from sklearn.isotonic import IsotonicRegression
    >>> X, y = make_regression(n_samples=10, n_features=1, random_state=41)
    >>> iso_reg = IsotonicRegression().fit(X.flatten(), y)

[rth]

BTW IsotonicRegression is currently the only estimator that has {'X_types': ['1darray']} estimator tags, where 1darray is never used in common tests, so they check nothing I imagine.

Though supporting 2darray with only n_features=1 in common tests might also be non trivial.

[amueller]

Right now any X_types that's not 2darray basically amounts to a "skip".
There's a PR somewhere with enforce_estimator_tags_X that ideally would enforce this tag.

[amueller]

#14705

[jnothman]

Unless there are obvious downsides we should probably make this change for consistency, allowing isotonic to take a column vector.

[amueller]

Agreed

[fujiaxiang]

Hi, can I work on this?

[fujiaxiang]

take