MiniBatchKMeans normalizes sample weights for each minibatch

[jeremiedbb]

In this line of the minibatch step of MiniBatchKMeans,

nearest_center, inertia = _labels_inertia(X, sample_weight, x_squared_norms, centers)

X is a minibatch, and internally, _labels_inertia will call _check_normalize_sample_weight which will normalize the sample weights according the minibatch and not to the full X.

It should be normalized once according to the full X at the beginning and that's all.

I'm working on it.

[jeremiedbb]

Not normalizing sample weights as discussed in #16594 will automatically fix that

[ogrisel]

@jeremiedbb it seems that we can close this now that #17848 has been merged. Do you agree?

[jeremiedbb]

Right, thanks