Random test failure in test_col_distribution_mask

[maxsch3]

    def test_col_distribution_mask(self):
        col_probs = [.5, .3, .2]
        cols = ['var1', 'var2', 'label']
        rct = BaseRandomCellTransform([.0, 1.], cols, col_probs)
        data = self.df.sample(10000, replace=True)
        mask = rct._make_mask(data)
        assert type(mask) == np.ndarray
        assert mask.shape == (data.shape[0], len(cols))
        # check if it is a proper one-hot encoding
        assert mask.sum() == data.shape[0]
        expected_counts = [5250, 3050, 1700]
        threshold = .0001
        # the counts do not make counts ideally to expected 5000, 3000, 2000
        c, p = chisquare(mask.sum(0), expected_counts)
        if p <= threshold:
            print(f'Error. looks like the column distribution {mask.sum(0)} is too far from expected '
                  f'{expected_counts}')
>       assert p > threshold
E       assert 1.8048484428487904e-06 > 0.0001

the printed message was:
Error. looks like the column distribution [5506. 2872. 1622.] is too far from expected [5250, 3050, 1700]

[maxsch3]

I will collect info from several failures like this to work out the solution