Random state parameter doomed to fail in IterativeStratification

[kamilc-bst]

Hi there,

First of all great job on stratification - it is already very useful in our recent project :)

I encountered small issue tho
I'm trying to use IterativeStratification and one it's parameters, random_state seems to be a trap

This code works fine:

from skmultilearn.model_selection import IterativeStratification
test_size = 0.2

stratifier = IterativeStratification(
    n_splits=2, order=2,
    sample_distribution_per_fold=[
        test_size, 1.0 - test_size],
)

train_indices, test_indices = next(
    stratifier.split(
        X=np.random.random((100,4)), 
        y=(np.random.random((100,4)) > 0.5).astype(int)    
    )
)

while this

from skmultilearn.model_selection import IterativeStratification
test_size = 0.2

stratifier = IterativeStratification(
    n_splits=2, order=2,
    sample_distribution_per_fold=[
        test_size, 1.0 - test_size],
    random_state = 42
)

train_indices, test_indices = next(
    stratifier.split(
        X=np.random.random((100,4)), 
        y=(np.random.random((100,4)) > 0.5).astype(int)    
    )
)

produces

ValueError: Setting a random_state has no effect since shuffle is False. You should leave random_state to its default (None), or set shuffle=True.

but shuffle is hardcoded as False in IterativeStratification super class call

scikit-multilearn/skmultilearn/model_selection/iterative_stratification.py

Line 184 in e6eabf0

shuffle=False,

[kamilc-bst]

that seems to be related to #234

[suchith-sixsense]

#248 gives the fix for IterativeStratification.

[erikhuck]

I was able to do a workaround using numpy.random.seed.