def test_next_batch_drop_last_false_1():
""" When 'drop_last' is False 'next_batch' should cycle through index. """
dsi = DatasetIndex(5)
left = []
right = list(np.concatenate([dsi.index, dsi.index]))
for length in [3, 3, 4]:
batch = dsi.next_batch(batch_size=length, n_epochs=2, drop_last=False)
left.extend(list(batch.index))
assert left == right
def test_next_batch_drop_last_false_2():
""" When 'drop_last' is False last batch of last epoch can have smaller length. """
dsi = DatasetIndex(5)
left = []
right = [2] * 7 + [
1
] # first seven batches have length of 2, last contains one item
for _ in range(8):
batch = dsi.next_batch(batch_size=2, n_epochs=3, drop_last=False)
left.append(len(batch))
assert left == right
def test_split_correctness():
""" Each element of 'index' is used.
Constants in 'shares' are such that test does not raise errors.
"""
dsi = DatasetIndex(5)
shares = .3 - np.random.random(3) * .05
dsi.split(shares=shares)
assert set(dsi.index) == (set(dsi.train.index)
| set(dsi.test.index)
| set(dsi.validation.index))
def test_next_batch_drop_last_true_2():
""" Order and contents of generated batches may differ at different epochs.
'shuffle' is True, so dropped indices are different at every epoch.
"""
dsi = DatasetIndex(5)
left = set()
right = set(dsi.index)
for _ in range(10):
batch = dsi.next_batch(batch_size=2,
n_epochs=None,
drop_last=True,
shuffle=True)
left = left | set(batch.index)
assert left == right
def test_next_batch_drop_last_true():
""" Order and contents of generated batches is same at every epoch.
'shuffle' is False, so dropped indices are always the same.
"""
dsi = DatasetIndex(5)
for _ in range(10):
batch_1 = dsi.next_batch(batch_size=2,
n_epochs=None,
drop_last=True,
shuffle=False)
batch_2 = dsi.next_batch(batch_size=2,
n_epochs=None,
drop_last=True,
shuffle=False)
assert (batch_1.index == dsi.index[:2]).all()
assert (batch_2.index == dsi.index[2:4]).all()
def test_calc_split_raise():
dsi = DatasetIndex(5)
with pytest.raises(ValueError):
dsi.calc_split(shares=[0.5, 0.5, 0.5])
with pytest.raises(ValueError):
dsi.calc_split(shares=[0.5, 0.5, 0.5, 0.5])
with pytest.raises(ValueError):
DatasetIndex(2).calc_split(shares=[0.5, 0.5, 0.5])
