def test_basic_cv(self):
c = self.make_config()
n_folds = random.randint(2,10)
repeat = 2
scores = models.cv(self.ds, c, n_folds, repeat)
self.assertAlmostEqual(scores.mean(), 0)
self.assertEqual(len(scores), n_folds*repeat)
def test_basic_cv_bad_valid(self):
# shouldnt matter
c = self.make_config(target='2a_plus_b_valid_diff')
n_folds = random.randint(2,10)
repeat = 2
scores = models.cv(self.ds, c, n_folds, repeat)
self.assertAlmostEqual(scores.mean(), 0)
self.assertEqual(len(scores), n_folds*repeat)
def test_basic_cv_no_relationship(self):
c = self.make_config(features=[
'c'
])
n_folds = random.randint(2,10)
repeat = 2
scores = models.cv(self.ds, c, n_folds, repeat)
self.assertTrue(3 < scores.mean() < 6)
self.assertEqual(len(scores), n_folds*repeat)
def test_blending(self):
pcnf = Configuration(
target='b',
features=[
'2a_plus_b_valid_diff',
'a'],
model= TestPredictor(linear_model.LinearRegression())
)
c = self.make_config(
target='b',
features=[
Predictions(pcnf)
])
scores = models.cv(self.ds, c)
self.assertEqual(pcnf.model.n_fit, 5)
print scores
self.assertTrue( 3 < scores.mean() < 5)