def test_make_k_folds_stratification_random(self):
"""Ensure stratification is preserved even with shuffling."""
for random_seed in np.random.randint(0, 1337, 100):
tts = TTS(shuffle=True, random_seed=random_seed)
for fold in tts.make_k_folds(self.y, n_folds=7, stratify=True):
np.testing.assert_allclose(np.sort(self.y[fold]),
np.array([1, 2, 3]))
def test_split_stratification_no_shuffle(self):
"""Ensure stratification is preserved if no shuffle."""
tts = TTS(shuffle=False)
train, test = tts.split(self.y, train_ratio=4. / 7., stratify=True)
assert np.count_nonzero(train == 1) ==\
np.count_nonzero(train == 2) == np.count_nonzero(train == 3)
assert np.count_nonzero(test == 1) == \
np.count_nonzero(test == 2) == np.count_nonzero(test == 3)
def test_make_k_folds_no_shuffle(self):
"""Ensure order is preserved if no shuffle (for different number of folds)."""
for n_folds in xrange(2, 100):
tts = TTS(shuffle=False)
folds = list(
tts.make_k_folds(self.y, n_folds=n_folds, stratify=False))
np.testing.assert_allclose(np.arange(len(self.y)),
np.concatenate(folds))
def test_split_stratification_random(self):
"""Ensure stratification is preserved even with shuffling."""
for random_seed in np.random.randint(0, 1337, 100):
tts = TTS(shuffle=True, random_seed=random_seed)
train, test = tts.split(self.y, train_ratio=4. / 7., stratify=True)
assert np.count_nonzero(train == 1) == \
np.count_nonzero(train == 2) == np.count_nonzero(train == 3)
assert np.count_nonzero(test == 1) == \
np.count_nonzero(test == 2) == np.count_nonzero(test == 3)
def test_split_no_shuffle(self):
"""Ensure order is preserved if no shuffle."""
tts = TTS(shuffle=False)
for train_ratio in np.arange(0.15, 0.96, 0.1):
train, test = tts.split(self.y,
train_ratio=train_ratio,
stratify=False)
np.testing.assert_allclose(
self.y, np.concatenate((self.y[train], self.y[test])))
def test_split_random_seed_2(self):
"""
Ensure different TTS instances with the same `random_seed` produce
the same outputs (for the same inputs).
"""
for train_ratio in (0.25, 0.5, 0.75):
for random_seed in np.random.randint(0, 1337, 25):
for stratify in (False, True):
for i in xrange(2):
if i == 0:
tts1 = TTS(shuffle=True, random_seed=random_seed)
train1, test1 = tts1.split(self.y,
train_ratio=train_ratio,
stratify=stratify)
if i == 1:
tts2 = TTS(shuffle=True, random_seed=random_seed)
train2, test2 = tts2.split(self.y,
train_ratio=train_ratio,
stratify=stratify)
np.testing.assert_allclose(train1, train1)
np.testing.assert_allclose(test1, test1)
def test_split_random_seed(self):
"""
Ensure for fixed (known) `random_seed` TTS instances produce
the same outputs (for the same inputs).
"""
for random_seed, stratify, (train_ref, test_ref) in (
(42, False, (np.array([6, 8, 9, 15, 7, 3, 17, 14, 11, 16]),
np.array([2, 19, 18, 1, 20, 10, 12, 4, 5, 0, 13]))),
(1337, False, (np.array([8, 20, 9, 18, 1, 4, 0, 7, 14, 16]),
np.array([17, 3, 15, 11, 5, 13, 2, 19, 6, 10,
12]))),
(42, True,
(np.array([12, 6, 4, 2, 3, 5, 1, 0, 9]),
np.array([20, 13, 16, 17, 18, 15, 8, 19, 10, 14, 7, 11]))),
(1337, True,
(np.array([6, 5, 1, 12, 0, 3, 2, 4, 9]),
np.array([14, 19, 17, 13, 10, 7, 16, 11, 8, 18, 15, 20]))),
):
tts = TTS(shuffle=True, random_seed=random_seed)
train, test = tts.split(self.y, train_ratio=0.5, stratify=stratify)
np.testing.assert_allclose(train_ref, train)
np.testing.assert_allclose(test_ref, test)
def test_make_k_folds_stratification_no_shuffle(self):
"""Ensure stratification is preserved if no shuffle."""
tts = TTS(shuffle=False)
for fold in tts.make_k_folds(self.y, n_folds=7, stratify=True):
np.testing.assert_allclose(np.sort(self.y[fold]),
np.array([1, 2, 3]))