def test_univariate_feature_screening(dim=(11, 12, 13), n_samples=10):
rng = np.random.RandomState(42)
mask = rng.rand(*dim) > 100. / np.prod(dim)
assert_true(mask.sum() >= 100.)
mask[dim[0] // 2, dim[1] // 3:, -dim[2] // 2:] = 1 # put spatial structure
n_features = mask.sum()
X = rng.randn(n_samples, n_features)
w = rng.randn(n_features)
w[rng.rand(n_features) > .8] = 0.
y = X.dot(w)
for is_classif in [True, False]:
X_, mask_, support_ = _univariate_feature_screening(
X, y, mask, is_classif, 20.)
n_features_ = support_.sum()
assert_equal(X_.shape[1], n_features_)
assert_equal(mask_.sum(), n_features_)
assert_true(n_features_ <= n_features)
def test_univariate_feature_screening(dim=(11, 12, 13), n_samples=10):
rng = np.random.RandomState(42)
mask = rng.rand(*dim) > 100. / np.prod(dim)
assert_true(mask.sum() >= 100.)
mask[dim[0] // 2, dim[1] // 3:, -dim[2] // 2:] = 1 # put spatial structure
n_features = mask.sum()
X = rng.randn(n_samples, n_features)
w = rng.randn(n_features)
w[rng.rand(n_features) > .8] = 0.
y = X.dot(w)
for is_classif in [True, False]:
X_, mask_, support_ = _univariate_feature_screening(
X, y, mask, is_classif, 20.)
n_features_ = support_.sum()
assert_equal(X_.shape[1], n_features_)
assert_equal(mask_.sum(), n_features_)
assert_true(n_features_ <= n_features)