def test_normalize_no_group(self):
# Fit with projector
pp = SCNormalizer(log_base=None)
data = pp(self.iris)
row_sums = data.X.sum(axis=1)
np.testing.assert_almost_equal(row_sums, row_sums[0])
def test_normalize_works_as_preprocessor(self):
pp = SCNormalizer(normalize_cells=False, log_base=2)
data = pp(self.iris)
data2 = self.iris.transform(data.domain)
np.testing.assert_almost_equal(data.X, data2.X)
np.testing.assert_array_equal(data.ids, data2.ids)
def test_normalize_bin(self):
BIN_THRESH = 3.0
pp = SCNormalizer(normalize_cells=False,
log_base=None,
bin_thresh=BIN_THRESH)
data = pp(self.iris)
expected_X = (self.iris.X > BIN_THRESH).astype(int)
np.testing.assert_almost_equal(data.X, expected_X)
def test_normalize_log(self):
LOG_BASE = 2.7
pp = SCNormalizer(normalize_cells=False, log_base=LOG_BASE)
data = pp(self.iris)
expected_X = np.log(1 + self.iris.X) / np.log(LOG_BASE)
np.testing.assert_almost_equal(data.X, expected_X)
def test_normalized_data_can_be_pickled(self):
pp = SCNormalizer(normalize_cells=False, log_base=2)
data = pp(self.iris)
data2 = pickle.loads(pickle.dumps(data))
np.testing.assert_almost_equal(data.X, data2.X)
np.testing.assert_array_equal(data.ids, data2.ids)
data3 = self.iris.transform(data2.domain)
np.testing.assert_almost_equal(data.X, data3.X)
np.testing.assert_array_equal(data.ids, data3.ids)
def test_normalize_with_group(self):
pp = SCNormalizer(equalize_var=self.iris.domain.class_var,
log_base=None)
data = pp(self.iris)
group1, group2, group3 = [self.iris.Y == i for i in range(3)]
med1 = np.median(data.X[group1].sum(axis=1))
med2 = np.median(data.X[group2].sum(axis=1))
med3 = np.median(data.X[group3].sum(axis=1))
# Group medians should be equal
self.assertAlmostEqual(med1, med2)
self.assertAlmostEqual(med2, med3)
def test_normalize_nans(self):
# Fit with projector
pp = SCNormalizer(log_base=None,
normalize_cells=True,
equalize_var="iris")
data = self.iris.copy()
data_nan = self.iris.copy()
for j in range(data_nan.X.shape[1]):
data.X[j, j] = 0
data_nan.X[j, j] = np.nan
data1 = pp(data)
data2 = pp(data_nan)
self.assertEqual(np.nansum(data2.X), np.sum(data1.X))
def commit(self):
log_base = self.log_base if self.log_check else None
library_var = None
if self.data is not None and \
self.normalize_cells and \
self.selected_attr in self.data.domain:
library_var = self.data.domain[self.selected_attr]
pp = SCNormalizer(equalize_var=library_var,
normalize_cells=self.normalize_cells,
log_base=log_base)
data = None
if self.data is not None:
data = pp(self.data)
self.Outputs.data.send(data)
self.Outputs.preprocessor.send(pp)
def update_preprocessors(self):
""" Update parameters of processors. """
log_base = self.log_base if self.log_check else None
library_var = None
selected_attr = self.attrs_model[self.selected_attr_index]
batch_link = self.LINK_FUNCTIONS[self.batch_link_index]
if self.data is not None and \
self.normalize_cells and \
selected_attr in self.data.domain:
library_var = self.data.domain[selected_attr]
self.pp = SCNormalizer(equalize_var=library_var,
normalize_cells=self.normalize_cells,
log_base=log_base)
self.pp_batch = SCBatchNormalizer(link=batch_link,
nonzero_only=batch_link == LINK_LOG,
batch_vars=self.batch_vars_selected)
def test_normalize_categorical(self):
pp = SCNormalizer()
dom = Domain([self.iris.domain.class_var])
data2 = self.iris.transform(dom)
norm_data = pp(data2)
np.testing.assert_array_equal(norm_data.X, data2.X)
