documentation of constraints in `isotonic_regression`

The docstring of isotonic_regression says https://github.com/scikit-learn/scikit-learn/blob/91badfab662ad1eafda4b60d078f563f010265cc/sklearn/isotonic.py#L84

Notice the equalities with respect to min and max.

However, the function does not actually force the smallest estimate to be y_min and the largest to be y_max. I believe that is the correct behavior; the estimates should be constrained, not fixed.

The docstring should therefore read

        subject to y_min <= y_[1] <= y_[2] ... <= y_[n] <= y_max

If this is changed, https://github.com/scikit-learn/scikit-learn/blob/91badfab662ad1eafda4b60d078f563f010265cc/sklearn/isotonic.py#L147 should be changed as well.

What the function does is to apply np.clip, i.e. first unconstrained estimates are produced and then everything outside of [y_min, y_max] is forced to these values. https://github.com/scikit-learn/scikit-learn/blob/91badfab662ad1eafda4b60d078f563f010265cc/sklearn/isotonic.py#L135 I am not sure that is the correct approach for constrained quadratic optimization (but also don’t know positively it to be incorrect). Tagging @bnaul and @agramfort since they seem to have introduced this line.

[NicolasHug]

This one was actually fixed in minus 5 days 😉

[NicolasHug]

Taken care of in #16234 (your review is welcome there!)