max_iter documentation seems wrong in LogisticRegression

[GaelVaroquaux]

It seems that the documentation for max_iter is wrong in the docstring of LogisticRegression: the documentation claims that the parameter is not passed to saga and liblinear.

max_iter is clearly passed to saga.
From a cursory look at the code, it seems that it is also passed to liblinear.

The documentation should be fixed, including checking what liblinear actually does, and maybe a test should be written.

[btel]

The max_iter parameter seems to be propagated all the way down to liblinear solver. I didn't check whether it's used internally by the solver, but I have no reason to believe that it's not the case. The same applies to saga solver.

The docs mention:

max_iter : int, optional (default=100)
Useful only for the newton-cg, sag and lbfgs solvers.
Maximum number of iterations taken for the solvers to converge.

scikit-learn/sklearn/linear_model/logistic.py

Line 1317 in b0ab289

max_iter : int, optional (default=100)

I guess liblinear and saga are not mentioned by omission.

[btel]

Here is a sample script, which confirms the case:

from sklearn.linear_model import LogisticRegression
import numpy as np

X = np.random.rand(100, 2)
y = X.sum(1) > 0.8

logreg = LogisticRegression(solver='saga', max_iter=1)
logreg.fit(X, y)

print("saga n iterations:", logreg.n_iter_)

logreg = LogisticRegression(solver='liblinear', max_iter=1)
logreg.fit(X, y)

print("liblinear iterations:", logreg.n_iter_)

which produces this output:

/home/bartosz/repos/libraries/scikit-learn/sklearn/linear_model/sag.py:334: ConvergenceWarning: The max_iter was reached which means the coef_ did not converge
  "the coef_ did not converge", ConvergenceWarning)
saga n iterations: [1]
/home/bartosz/repos/libraries/scikit-learn/sklearn/svm/base.py:927: ConvergenceWarning: Liblinear failed to converge, increase the number of iterations.
  "the number of iterations.", ConvergenceWarning)
liblinear iterations: [1]