def test_new_duplicate_label(self):
c1 = DummyClassifier()
c2 = DummyClassifier()
self.assertRaisesRegex(ValueError,
"Duplicate classifier label 'c'",
ClassifierCollection, {'c': c1},
c=c2)
def test_classify_subset(self):
ccol = ClassifierCollection({
'subjectA': DummyClassifier(),
'subjectB': DummyClassifier(),
})
classifierB = ccol._label_to_classifier['subjectB']
classifierB.classify_one_element = mock.Mock()
d_v = [0, 1, 2, 3, 4]
d = DescriptorMemoryElement('memory', '0')
d.set_vector(d_v)
result = ccol.classify(d, labels=['subjectA'])
# Should contain one entry for each requested classifier.
self.assertEqual(len(result), 1)
self.assertIn('subjectA', result)
self.assertNotIn('subjectB', result)
classifierB.classify_one_element.assert_not_called()
# Each key should map to a classification element (memory in this case
# because we're using the default factory)
self.assertIsInstance(result['subjectA'], MemoryClassificationElement)
# We know the dummy classifier outputs "classifications" in a
# deterministic way: class label is descriptor UUID and classification
# value is its vector as a list.
self.assertDictEqual(result['subjectA'].get_classification(),
{'test': 0})
def test_classify_missing_label(self):
ccol = ClassifierCollection({
'subjectA': DummyClassifier(),
'subjectB': DummyClassifier(),
})
d_v = [0, 1, 2, 3, 4]
d = DescriptorMemoryElement('memory', '0')
d.set_vector(d_v)
# Should throw a MissingLabelError
with self.assertRaises(MissingLabelError) as cm:
ccol.classify(d, labels=['subjectC'])
self.assertSetEqual(cm.exception.labels, {'subjectC'})
# Should throw a MissingLabelError
with self.assertRaises(MissingLabelError) as cm:
ccol.classify(d, labels=['subjectA', 'subjectC'])
self.assertSetEqual(cm.exception.labels, {'subjectC'})
# Should throw a MissingLabelError
with self.assertRaises(MissingLabelError) as cm:
ccol.classify(d, labels=['subjectC', 'subjectD'])
self.assertSetEqual(cm.exception.labels, {'subjectC', 'subjectD'})
# Should throw a MissingLabelError
with self.assertRaises(MissingLabelError) as cm:
ccol.classify(d, labels=['subjectA', 'subjectC', 'subjectD'])
self.assertSetEqual(cm.exception.labels, {'subjectC', 'subjectD'})
def test_classify(self):
ccol = ClassifierCollection({
'subjectA': DummyClassifier(),
'subjectB': DummyClassifier(),
})
d_v = [0, 1, 2, 3, 4]
d = DescriptorMemoryElement('memory', '0')
d.set_vector(d_v)
result = ccol.classify(d)
# Should contain one entry for each configured classifier.
self.assertEqual(len(result), 2)
self.assertIn('subjectA', result)
self.assertIn('subjectB', result)
# Each key should map to a classification element (memory in this case
# because we're using the default factory)
self.assertIsInstance(result['subjectA'], MemoryClassificationElement)
self.assertIsInstance(result['subjectB'], MemoryClassificationElement)
# We know the dummy classifier outputs "classifications" in a
# deterministic way: class label is "test" and classification
# value is the index of the descriptor .
self.assertDictEqual(result['subjectA'].get_classification(),
{'test': 0})
self.assertDictEqual(result['subjectB'].get_classification(),
{'test': 0})
def test_add_classifier_duplicate_label(self):
ccol = ClassifierCollection(a=DummyClassifier())
self.assertRaisesRegexp(
ValueError,
"Duplicate label provided: 'a'",
ccol.add_classifier,
'a', DummyClassifier()
)
def test_labels(self):
ccol = ClassifierCollection(
classifiers={
'b': DummyClassifier(),
},
a=DummyClassifier(),
label2=DummyClassifier(),
)
self.assertEqual(ccol.labels(), {'a', 'b', 'label2'})
def test_size_len(self):
ccol = ClassifierCollection()
self.assertEqual(ccol.size(), 0)
self.assertEqual(len(ccol), 0)
ccol = ClassifierCollection(
a=DummyClassifier(),
b=DummyClassifier(),
)
self.assertEqual(ccol.size(), 2)
self.assertEqual(len(ccol), 2)
def test_get_config_with_stuff(self):
c1 = DummyClassifier()
c2 = DummyClassifier()
ccol = ClassifierCollection({'a': c1}, b=c2)
# dummy returns {} config.
self.assertEqual(
ccol.get_config(),
{
'a': {'DummyClassifier': {}, 'type': 'DummyClassifier'},
'b': {'DummyClassifier': {}, 'type': 'DummyClassifier'},
}
)
def test_remove_classifier_bad_label(self):
c = DummyClassifier()
ccol = ClassifierCollection(a=c)
self.assertRaises(
KeyError,
ccol.remove_classifier, 'b'
)
def test_add_classifier(self):
ccol = ClassifierCollection()
self.assertEqual(ccol.size(), 0)
c = DummyClassifier()
ccol.add_classifier('label', c)
self.assertEqual(ccol.size(), 1)
self.assertEqual(ccol._label_to_classifier['label'], c)
def test_classify_arrays(self):
""" Test invoking `classify_arrays` in a valid manner. """
# Use some dummy classifiers that
ccol = ClassifierCollection({
'subjectA': DummyClassifier(),
'subjectB': DummyClassifier(),
})
dmat = np.asarray([[0, 1, 2, 3, 4], [5, 6, 7, 8, 9]])
result = ccol.classify_arrays(dmat)
# Should contain one entry for each configured classifier.
assert len(result) == 2
assert 'subjectA' in result
assert 'subjectB' in result
# Each key should map to a list of dictionaries mapping classifier
# labels to confidence values. The DummyClassify maps input descriptor
# first values as the "confidence" for simplicity. Should be in order
# of input vectors.
assert result['subjectA'] == [{'test': 0}, {'test': 5}]
assert result['subjectB'] == [{'test': 0}, {'test': 5}]
def test_get_config_with_stuff(self):
c1 = DummyClassifier()
c2 = DummyClassifier()
ccol = ClassifierCollection({'a': c1}, b=c2)
# dummy returns {} config.
self.assertEqual(
ccol.get_config(), {
'a': {
'tests.algorithms.classifier.test_ClassifierAbstract'
'.DummyClassifier': {},
'type':
'tests.algorithms.classifier'
'.test_ClassifierAbstract.DummyClassifier'
},
'b': {
'tests.algorithms.classifier.test_ClassifierAbstract'
'.DummyClassifier': {},
'type':
'tests.algorithms.classifier'
'.test_ClassifierAbstract.DummyClassifier'
}
})
def test_classify_empty_subset(self):
ccol = ClassifierCollection({
'subjectA': DummyClassifier(),
'subjectB': DummyClassifier(),
})
classifierA = ccol._label_to_classifier['subjectA']
classifierA.classify_one_element = mock.Mock()
classifierB = ccol._label_to_classifier['subjectB']
classifierB.classify_one_element = mock.Mock()
d_v = [0, 1, 2, 3, 4]
d = DescriptorMemoryElement('memory', '0')
d.set_vector(d_v)
result = ccol.classify(d, labels=[])
# Should contain no entries.
self.assertEqual(len(result), 0)
self.assertNotIn('subjectA', result)
classifierA.classify_one_element.assert_not_called()
self.assertNotIn('subjectB', result)
classifierB.classify_one_element.assert_not_called()
def test_remove_classifier(self):
c = DummyClassifier()
ccol = ClassifierCollection(a=c)
ccol.remove_classifier('a')
self.assertEqual(ccol._label_to_classifier, {})
def test_new_both_pos_and_kwds(self):
c1 = DummyClassifier()
c2 = DummyClassifier()
ccol = ClassifierCollection({'a': c1}, b=c2)
self.assertEqual(ccol._label_to_classifier,
{'a': c1, 'b': c2})
def test_new_kwargs(self):
c = DummyClassifier()
ccol = ClassifierCollection(a_label=c)
self.assertEqual(ccol._label_to_classifier, {'a_label': c})
def test_new_positional(self):
c = DummyClassifier()
ccol = ClassifierCollection(classifiers={'a label': c})
self.assertEqual(ccol._label_to_classifier, {'a label': c})
def test_get_classifier(self):
c = DummyClassifier()
ccol = ClassifierCollection(a=c)
self.assertEqual(ccol.get_classifier('a'), c)
