def test_remove_empty_table_empty(self):
def make_bad(v, i, m):
return np.zeros(len(v))
diff_table = self.table.copy()
diff_table.transform(make_bad, inplace=True)
with self.assertRaisesRegex(
ValueError,
"All samples / features in matched table are empty."):
remove_empty_samples_and_features(diff_table, self.sm)
def test_remove_empty_table_empty_and_ordination_funky(self):
# Even if the ordination contains empty samples (as is the case for
# self.pcoa), the table being completely empty should still take
# precedence as an error. (If *both* errors are present for a dataset,
# then I recommend consulting a priest.)
def make_bad(v, i, m):
return np.zeros(len(v))
diff_table = self.table.copy()
diff_table.transform(make_bad, inplace=True)
with self.assertRaisesRegex(
ValueError,
"All samples / features in matched table are empty."):
remove_empty_samples_and_features(diff_table, self.sm, self.pcoa)
def test_remove_empty_with_empty_feature_in_ordination(self):
bad_feature_pcoa = skbio.OrdinationResults(
'PCoA',
'Principal Coordinate Analysis',
self.eigvals,
self.samples_df.drop(labels="Sample4", axis="index"),
features=self.features_df,
proportion_explained=self.proportion_explained)
with self.assertRaisesRegex(
ValueError,
(r"The ordination contains features that are empty \(i.e. all "
r"0s\) in the table. Problematic feature IDs: e")):
remove_empty_samples_and_features(self.table, self.sm,
bad_feature_pcoa)
def test_remove_empty_with_multiple_empty_samples_in_ordination(self):
def make_bad(v, i, m):
if i == 'Sample1':
return np.zeros(len(v))
else:
return v
bad_table = self.table.copy()
bad_table.transform(make_bad, inplace=True)
with self.assertRaisesRegex(
ValueError,
(r"The ordination contains samples that are empty \(i.e. all "
r"0s\) in the table. Problematic sample IDs: Sample1, Sample4")):
remove_empty_samples_and_features(bad_table, self.sm, self.pcoa)
def test_remove_empty_1_empty_sample_and_feature(self, mock_stdout):
ft, fsm = remove_empty_samples_and_features(self.table, self.sm)
self.assertEqual(ft, self.table_ef)
assert_frame_equal(fsm, self.sm_ef)
self.assertEqual(
mock_stdout.getvalue(),
"Removed 1 empty sample(s).\nRemoved 1 empty feature(s).\n")
def test_remove_empty_with_empty_sample_and_feature_in_ordination(self):
# Checks behavior when both an empty sample and an empty feature are in
# the ordination. Currently the code is structured so that empty sample
# errors take precedence over empty feature errors -- I imagine this
# will be the more common of the two scenarios, which is partially why
# I went with this. But this is probably a rare edge case anyway.
extremely_funky_pcoa = skbio.OrdinationResults(
'PCoA',
'Principal Coordinate Analysis',
self.eigvals,
self.samples_df,
features=self.features_df,
proportion_explained=self.proportion_explained)
with self.assertRaisesRegex(
ValueError,
(r"The ordination contains samples that are empty \(i.e. all "
r"0s\) in the table. Problematic sample IDs: Sample4")):
remove_empty_samples_and_features(self.table, self.sm,
extremely_funky_pcoa)
def _validate_and_match_data(self, ignore_missing_samples,
filter_extra_samples,
filter_missing_features,
shear_to_table,
shear_to_feature_metadata):
if self.is_community_plot:
self.table, self.samples, self.tip_md, self.int_md = match_inputs(
self.tree, self.table, self.samples, self.features,
self.ordination, ignore_missing_samples, filter_extra_samples,
filter_missing_features
)
# Remove empty samples and features from the table (and remove the
# removed samples from the sample metadata). We also pass in the
# ordination, if present, to this function -- so we can throw an
# error if the ordination actually contains these empty
# samples/features.
#
# We purposefully do this removal *after* matching (so we know the
# data inputs match up) and *before* shearing (so empty features
# in the table are no longer included as tips in the tree).
self.table, self.samples = remove_empty_samples_and_features(
self.table, self.samples, self.ordination
)
# remove unobserved features from the phylogeny (shear the tree)
if shear_to_table:
features = set(self.table.ids(axis='observation'))
self.tree = self.tree.shear(features)
# Remove features in the feature metadata that are no longer
# present in the tree, due to being shorn off
if self.tip_md is not None or self.int_md is not None:
# (Technically they should always both be None or both be
# DataFrames -- there's no in-between)
self.tip_md, self.int_md = filter_feature_metadata_to_tree(
self.tip_md, self.int_md, self.tree
)
else:
if shear_to_feature_metadata:
features = set(self.features.index)
all_tips = set(bp_tree_tips(self.tree))
# check that feature metadata contains at least 1 tip
if not features.intersection(all_tips):
raise ValueError(
"Cannot shear tree to feature metadata: no tips in "
"the tree are present in the feature metadata."
)
self.tree = self.tree.shear(features)
self.tip_md, self.int_md = match_tree_and_feature_metadata(
self.tree, self.features
)
validate_tree(self.tree)
def test_remove_empty_nothing_to_remove_with_ordination(self, mock_stdout):
good_pcoa = skbio.OrdinationResults(
'PCoA',
'Principal Coordinate Analysis',
self.eigvals,
self.samples_df.drop(labels="Sample4", axis="index"),
features=self.features_df.drop(labels="e", axis="index"),
proportion_explained=self.proportion_explained)
ft, fsm = remove_empty_samples_and_features(self.table_ef, self.sm_ef,
good_pcoa)
self.assertEqual(ft, self.table_ef)
assert_frame_equal(fsm, self.sm_ef)
self.assertEqual(mock_stdout.getvalue(), "")
def test_remove_empty_nothing_to_remove(self, mock_stdout):
ft, fsm = remove_empty_samples_and_features(self.table_ef, self.sm_ef)
self.assertEqual(ft, self.table_ef)
assert_frame_equal(fsm, self.sm_ef)
self.assertEqual(mock_stdout.getvalue(), "")
def test_remove_empty_with_empty_sample_in_ordination(self):
with self.assertRaisesRegex(
ValueError,
(r"The ordination contains samples that are empty \(i.e. all "
r"0s\) in the table. Problematic sample IDs: Sample4")):
remove_empty_samples_and_features(self.table, self.sm, self.pcoa)