def test_init_params(self):
"""Tests that KFold() raises errors on invalid parameters"""
with self.assertRaises(ValueError):
KFold(n_splits=2.5)
with self.assertRaises(ValueError):
KFold(n_splits=1)
with self.assertRaises(TypeError):
KFold(shuffle=None)
def test_split_no_shuffle(self):
"""Tests KFold.split() method with shuffle=False"""
x_np = np.random.rand(1000, 3)
y_np = np.arange(1000)
x = ds.array(x_np, (111, 3))
y = ds.array(y_np[:, np.newaxis], (111, 1))
cv = KFold(shuffle=False)
n_splits = 0
for train_ds, test_ds in cv.split(x, y):
len_x_test = test_ds[0].shape[0]
self.assertEqual(len_x_test, 200)
n_splits += 1
self.assertEqual(cv.get_n_splits(), n_splits)
def test_split_no_shuffle_uneven_folds(self):
"""Tests KFold.split() method with shuffle=False and uneven folds"""
x_np = np.random.rand(1000, 3)
y_np = np.arange(1000)
x = ds.array(x_np, (334, 3))
y = ds.array(y_np[:, np.newaxis], (334, 1))
cv = KFold(n_splits=3, shuffle=False)
n_splits = 0
for train_ds, test_ds in cv.split(x, y):
len_x_test = test_ds[0].shape[0]
self.assertTrue(len_x_test == 333 or len_x_test == 334,
'Fold size is ' + str(len_x_test) +
' and should be 333 or 334.')
n_splits += 1
self.assertEqual(cv.get_n_splits(), n_splits)
self.assertEqual(3, n_splits)
def test_split_single_subset_no_shuffle_uneven_folds(self):
"""Tests KFold.split() from single subset, shuffle=False and uneven"""
x_np = np.random.rand(1000, 3)
y_np = np.arange(1000)
x = ds.array(x_np, (1000, 3))
y = ds.array(y_np[:, np.newaxis], (1000, 1))
cv = KFold(n_splits=6, shuffle=False)
n_splits = 0
for train_ds, test_ds in cv.split(x, y):
len_x_train = train_ds[0].shape[0]
len_y_train = train_ds[1].shape[0]
self.assertEquals(len_x_train, len_y_train)
len_x_test = test_ds[0].shape[0]
len_y_test = test_ds[1].shape[0]
self.assertEquals(len_x_test, len_y_test)
self.assertEquals(len_x_train + len_x_test, 1000)
self.assertTrue(len_x_test == 166 or len_x_test == 167,
'Fold size is ' + str(len_x_test) +
' but should be 166 or 167.')
n_splits += 1
self.assertEqual(cv.get_n_splits(), n_splits)
def test_cv_class(self):
"""Tests GridSearchCV with a class cv parameter."""
x_np, y_np = datasets.load_iris(return_X_y=True)
x = ds.array(x_np, (30, 4))
y = ds.array(y_np[:, np.newaxis], (30, 1))
rf = RandomForestClassifier()
param_grid = {'n_estimators': (2, 4)}
searcher = GridSearchCV(rf, param_grid, cv=KFold(4))
searcher.fit(x, y)
self.assertTrue(hasattr(searcher, 'cv_results_'))
self.assertTrue(hasattr(searcher, 'best_estimator_'))
self.assertTrue(hasattr(searcher, 'best_score_'))
self.assertTrue(hasattr(searcher, 'best_params_'))
self.assertTrue(hasattr(searcher, 'best_index_'))
self.assertTrue(hasattr(searcher, 'scorer_'))