class TolerantEvaluatorFitTransformTests(TestCase, TolerantFittingCases):
# We try to mimic most of the cases seen on Fit tests, and check that
# they are working equivalently with fit_transform
fit_method_name = 'fit_transform'
def test_sample_is_excluded_if_any_feature_fails_when_evaluating_it(self):
self.ev = TolerantFeatureEvaluator(
[DescriptionFeature, EntireSampleFeature])
self.ev.FEATURE_STRICT_UNTIL = 0
self.ev.FEATURE_MAX_ERRORS_ALLOWED = len(SAMPLES) + 1 # dont exclude
samples = SAMPLES[:]
nodescription = {'nodescription': u'this sample has no description'}
samples.append(nodescription)
result = self.ev.fit_transform(samples)
self.assertTrue(len(list(result)) < len(samples))
# EntireSampleFeature is the last, so is the last value per tuple
self.assertNotIn(nodescription, [r[-1] for r in result])
def test_if_a_feature_is_excluded_all_results_doesnt_include_it(self):
# This means: if a Feature evaluated fine for some samples until it was
# excluded, once we decided to exclude it, we must make sure that
# previous samples for which this feature was evaluated, are now
# striped out of those evaluations
self.ev = TolerantFeatureEvaluator([DescriptionFeature, DumbFeatureA])
self.ev.FEATURE_STRICT_UNTIL = 0
self.ev.FEATURE_MAX_ERRORS_ALLOWED = 0 # No feature failure tolerated
result = self.ev.fit_transform(SAMPLES + [{'nodescription': u'tada!'}])
# Check that there are results. Otherwise, next loop is dumb
self.assertTrue(result)
for r in result:
self.assertEqual(len(r), 1) # only one value per sample
self.assertEqual(r[0], 'a') # Remember DumbFeatureA returns 'a'
def test_when_feature_is_excluded_discarded_samples_are_reevaluated(self):
self.ev = TolerantFeatureEvaluator(
[DescriptionFeature, DumbFeatureA, EntireSampleFeature])
self.ev.FEATURE_MAX_ERRORS_ALLOWED = 0 # No feature failure tolerated
samples = SAMPLES[:]
nodescription = {'nodescription': u'this sample has no description'}
samples.append(nodescription)
result = list(self.ev.fit_transform(samples))
self.assertEqual(len(samples), len(result))
# EntireSampleFeature is the last, so is the last value per tuple
self.assertIn(nodescription, [r[-1] for r in result])
def test_consumable_is_consumed_only_once(self):
samples = (s for s in SAMPLES) # can be consumed once only
self.ev = TolerantFeatureEvaluator([EntireSampleFeature])
result = list(self.ev.fit_transform(samples))
self.assertEqual(len(SAMPLES), len(result))
class TolerantEvaluatorFitTransformTests(TestCase, TolerantFittingCases):
# We try to mimic most of the cases seen on Fit tests, and check that
# they are working equivalently with fit_transform
fit_method_name = 'fit_transform'
def test_sample_is_excluded_if_any_feature_fails_when_evaluating_it(self):
self.ev = TolerantFeatureEvaluator([DescriptionFeature,
EntireSampleFeature])
self.ev.FEATURE_STRICT_UNTIL = 0
self.ev.FEATURE_MAX_ERRORS_ALLOWED = len(SAMPLES) + 1 # dont exclude
samples = SAMPLES[:]
nodescription = {'nodescription': u'this sample has no description'}
samples.append(nodescription)
result = self.ev.fit_transform(samples)
self.assertTrue(len(list(result)) < len(samples))
# EntireSampleFeature is the last, so is the last value per tuple
self.assertNotIn(nodescription, [r[-1] for r in result])
def test_if_a_feature_is_excluded_all_results_doesnt_include_it(self):
# This means: if a Feature evaluated fine for some samples until it was
# excluded, once we decided to exclude it, we must make sure that
# previous samples for which this feature was evaluated, are now
# striped out of those evaluations
self.ev = TolerantFeatureEvaluator([DescriptionFeature, DumbFeatureA])
self.ev.FEATURE_STRICT_UNTIL = 0
self.ev.FEATURE_MAX_ERRORS_ALLOWED = 0 # No feature failure tolerated
result = self.ev.fit_transform(SAMPLES + [{'nodescription': u'tada!'}])
# Check that there are results. Otherwise, next loop is dumb
self.assertTrue(result)
for r in result:
self.assertEqual(len(r), 1) # only one value per sample
self.assertEqual(r[0], 'a') # Remember DumbFeatureA returns 'a'
def test_when_feature_is_excluded_discarded_samples_are_reevaluated(self):
self.ev = TolerantFeatureEvaluator([DescriptionFeature, DumbFeatureA,
EntireSampleFeature])
self.ev.FEATURE_MAX_ERRORS_ALLOWED = 0 # No feature failure tolerated
samples = SAMPLES[:]
nodescription = {'nodescription': u'this sample has no description'}
samples.append(nodescription)
result = list(self.ev.fit_transform(samples))
self.assertEqual(len(samples), len(result))
# EntireSampleFeature is the last, so is the last value per tuple
self.assertIn(nodescription, [r[-1] for r in result])
def test_consumable_is_consumed_only_once(self):
samples = (s for s in SAMPLES) # can be consumed once only
self.ev = TolerantFeatureEvaluator([EntireSampleFeature])
result = list(self.ev.fit_transform(samples))
self.assertEqual(len(SAMPLES), len(result))
