[Python] Accept numpy generators as random_state

[david-cortes]

This PR modifies the scikit-learn interface to accept numpy generators as possible inputs for random_state. Generators are now the recommended mechanism for drawing random numbers in numpy, while the previous RandomState is deprecated. Lots of other software like SciPy have moved towards the new Generator class and allow passing a random_state as either int/RandomState/Generator (example).

[trivialfis]

Thank you for the work on rng!

[trivialfis]

Hi, could you please help take a look into the CI errors? Feel free to ping me if there's anything I can help.

[david-cortes]

Hi, could you please help take a look into the CI errors? Feel free to ping me if there's anything I can help.

This is the message I managed to find:

XFAIL tests/python/test_with_modin.py::TestModin::test_modin

The PR didn't touch anything that would deal with modin so don't know the reason for the failure.

[trivialfis]

This is the message I managed to find:

Hmm, not sure where did you see the modin error. Here are the ones I see on github action:

https://github.com/dmlc/xgboost/actions/runs/6719062791/job/18259950760?pr=9743
https://github.com/dmlc/xgboost/actions/runs/6719062791/job/18259951139?pr=9743
https://github.com/dmlc/xgboost/actions/runs/6719062791/job/18259950609?pr=9743

 ================================== FAILURES ===================================
__________________________ test_sklearn_random_state __________________________

    def test_sklearn_random_state():
        clf = xgb.XGBClassifier(random_state=402)
        assert clf.get_xgb_params()['random_state'] == 402
    
        clf = xgb.XGBClassifier(random_state=401)
        assert clf.get_xgb_params()['random_state'] == 401
    
        random_state = np.random.RandomState(seed=403)
        clf = xgb.XGBClassifier(random_state=random_state)
        assert isinstance(clf.get_xgb_params()['random_state'], int)
    
        random_state = np.random.default_rng(seed=404)
        clf = xgb.XGBClassifier(random_state=random_state)
>       assert isinstance(clf.get_xgb_params()['random_state'], int)
E       assert False
E        +  where False = isinstance(1983313941, int)

[david-cortes]

Thanks for the hints. Fixed now.