def test_minibatch_idx_iterator():
'''
Testing the minibatches of idxs are made correctly
minibatch_idx_iterator(Y, minibatch_size, randomise, balanced)
'''
# simplest case - classes are already balanced and minibatch_size == len(Y)
Y = [0, 0, 1, 1, 2, 2]
ans = list(mbg.minibatch_idx_iterator(
Y, 6, randomise=False, balanced=False))
# ans is a list of lists, the first of which should contain everything
assert len(ans) == 1
assert set(ans[0]) == {0, 1, 2, 3, 4, 5}
# now with randomisation
ans = list(mbg.minibatch_idx_iterator(Y, 6, randomise=True, balanced=False))
assert len(ans) == 1
assert set(ans[0]) == {0, 1, 2, 3, 4, 5}
def test_imbalanced_minibatch_idx_iterator():
# now with imbalanced classes
Y = [0, 1, 1, 2, 2, 2, 3, 3, 3, 3]
ans = list(mbg.minibatch_idx_iterator(
Y, 10, randomise=False, balanced=False))
assert len(ans) == 1
assert set(ans[0]) == {0, 1, 2, 3, 4, 5, 6, 7, 8, 9}
ans = list(mbg.minibatch_idx_iterator(
Y, 10, randomise=False, balanced=True))
# simple minibatch check
check_minibatches(ans, 10)
# classes should be approximately balanced
for an in ans:
class_counts = np.bincount([Y[xx] for xx in an])
assert class_counts.max() - class_counts.min() <= 1
# all idxs should be returned at least once
all_idxs = [yy for xx in ans for yy in xx]
assert set(all_idxs) == {0, 1, 2, 3, 4, 5, 6, 7, 8, 9}