def test_vectors_and_arrays(self):
X = np.random.normal(size=(10, 3))
Y = np.random.normal(size=(10, 2))
model = SelectiveRegularization(unpenalized_inds=[0],
penalized_model=Ridge(),
fit_intercept=True)
self.assertEqual(model.fit(X, Y).coef_.shape, (2, 3))
self.assertEqual(model.fit(X, Y[:, 0]).coef_.shape, (3, ))
def test_can_use_sample_weights(self):
n = 100 + np.random.choice(500)
d = 5 + np.random.choice(20)
n_inds = np.random.choice(np.arange(2, d - 2))
inds = np.random.choice(d, n_inds, replace=False)
alpha = np.random.uniform(0.5, 1.5)
X = np.random.normal(size=(n, d))
y = np.random.normal(size=(n, ))
sample_weight = np.random.choice([1, 2], n)
# create an extra copy of rows with weight 2
X_aug = X[sample_weight == 2, :]
y_aug = y[sample_weight == 2]
model = SelectiveRegularization(unpenalized_inds=inds,
penalized_model=Ridge(),
fit_intercept=True)
coef = model.fit(X, y, sample_weight=sample_weight).coef_
coef2 = model.fit(np.vstack((X, X_aug)), np.concatenate(
(y, y_aug))).coef_
np.testing.assert_allclose(coef, coef2)
def test_can_pass_through_attributes(self):
X = np.random.normal(size=(10, 3))
y = np.random.normal(size=(10, ))
model = SelectiveRegularization(unpenalized_inds=[0],
penalized_model=LassoCV(),
fit_intercept=True)
# _penalized_inds is only set during fitting
with self.assertRaises(AttributeError):
inds = model._penalized_inds
# cv exists on penalized model
old_cv = model.cv
model.cv = 2
model.fit(X, y)
# now we can access _penalized_inds
assert np.array_equal(model._penalized_inds, [1, 2])
# check that we can read the cv attribute back out from the underlying model
assert model.cv == 2