def test_can_preserve_more_components(self):
selector = fs.GMMSelector("mean", n_candidates=-1).fit(self.data)
assert accuracy_score(self.labels >= 2, selector.selected_) >= 0.99
selector = fs.GMMSelector("mean", n_candidates=2).fit(self.data)
assert accuracy_score(self.labels >= 2, selector.selected_) >= 0.99
selector = fs.GMMSelector("mean", n_candidates=1).fit(self.data)
assert accuracy_score(self.labels >= 1, selector.selected_) >= 0.99
selector = fs.GMMSelector("mean", n_candidates=1, preserve_high=False)
selector.fit(self.data)
assert accuracy_score(self.labels < 3, selector.selected_) >= 0.99
def test_can_preserve_more_components(self):
selector = fs.GMMSelector('mean', n_candidates=-1).fit(self.data)
self.assertGreaterEqual(
accuracy_score(self.labels >= 2, selector.selected_), 0.99)
selector = fs.GMMSelector('mean', n_candidates=2).fit(self.data)
self.assertGreaterEqual(
accuracy_score(self.labels >= 2, selector.selected_), 0.99)
selector = fs.GMMSelector('mean', n_candidates=1).fit(self.data)
self.assertGreaterEqual(
accuracy_score(self.labels >= 1, selector.selected_), 0.99)
selector = fs.GMMSelector('mean', n_candidates=1, preserve_high=False)
selector.fit(self.data)
self.assertGreaterEqual(
accuracy_score(self.labels < 3, selector.selected_), 0.99)
def test_preserves_min_features_precisely_or_rate(self):
selector = fs.GMMSelector('mean', min_features=50).fit(self.data)
self.assertGreaterEqual(selector.selected_.sum(), 50)
selector = fs.GMMSelector('mean', min_features_rate=0.5).fit(self.data)
self.assertGreaterEqual(selector.selected_.sum(),
0.5 * self.labels.size)
def test_works_for_mean_and_var_only(self):
fs.GMMSelector('mean')
fs.GMMSelector('var')
with self.assertRaises(ValueError):
fs.GMMSelector('yolo')
def test_fails_log_with_negative_features(self):
self.data[0, 1] = -1
selector = fs.GMMSelector('mean', use_log=True)
with self.assertRaises(ValueError):
selector.fit(self.data)
def test_that_0_candidates_preserves_all_the_features(self):
selector = fs.GMMSelector('mean', n_candidates=0).fit(self.data)
npt.assert_array_equal(selector.selected_, True)
def test_selects_low_component_when_set(self):
selector = fs.GMMSelector('mean', preserve_high=False).fit(self.data)
npt.assert_array_equal(selector.selected_,
self.labels == self.labels.min())
def test_selects_high_component_by_default(self):
selector = fs.GMMSelector('mean').fit(self.data)
npt.assert_array_equal(selector.selected_,
self.labels == self.labels.max())
def test_preserves_min_features_precisely_or_rate(self):
selector = fs.GMMSelector("mean", min_features=50).fit(self.data)
assert selector.selected_.sum() >= 50
selector = fs.GMMSelector("mean", min_features_rate=0.5).fit(self.data)
assert selector.selected_.sum() >= 0.5 * self.labels.size
def test_skips_neutral(self):
data = self.labels.reshape(-1, 1).astype(float)
m1 = fs.GMMSelector("mean", neutral=0.0).fit(data).vals_
m2 = data[30:].mean()
assert m1 == m2
def test_works_for_selected_metrics_only(self):
fs.GMMSelector("mean")
fs.GMMSelector("var")
fs.GMMSelector("cv")
with pytest.raises(ValueError):
fs.GMMSelector("yolo")
def test_fails_log_with_negative_features(self):
self.data[0, 1] = -1
selector = fs.GMMSelector("mean", use_log=True)
with pytest.raises(ValueError):
selector.fit(self.data)
