def test_calculate_feature_importances_ensemble(self):
estimator = Pipeline([('dv', DictVectorizer()),
('est', RandomForestClassifier(n_estimators=10))
])
estimator.fit(_extract_features(self.features),
self.targets.values.ravel())
fi = _calculate_feature_importances(estimator)
self.assertEqual(sorted(self.exp_rf), sorted(fi))
def test_null_feature_importance(self):
exp = pd.DataFrame([1, 1, 1],
index=['o1', 'o2', 'o3'],
columns=['importance'])
exp.index.name = 'feature'
tab = biom.Table(np.array([[1., 2., 3.], [3., 2., 1.], [7., 6., 9.]]),
['o1', 'o2', 'o3'], ['s1', 's2', 's3'])
tab = _extract_features(tab)
pdt.assert_frame_equal(_null_feature_importance(tab), exp)
def test_extract_features(self):
table = self.table_ecam_fp
dicts = _extract_features(table)
dv = DictVectorizer()
dv.fit(dicts)
features = table.ids('observation')
self.assertEqual(set(dv.get_feature_names()), set(features))
self.assertEqual(len(dicts), len(table.ids()))
for dict_row, (table_row, _, _) in zip(dicts, table.iter()):
for feature, count in zip(features, table_row):
if count == 0:
self.assertTrue(feature not in dict_row)
else:
self.assertEqual(dict_row[feature], count)
def test_calculate_feature_importances_svm(self):
estimator = Pipeline([('dv', DictVectorizer()), ('est', LinearSVC())])
estimator.fit(_extract_features(self.features),
self.targets.values.ravel())
fi = _calculate_feature_importances(estimator)
self.assertEqual(sorted(self.exp_lsvm), sorted(fi))
def biom_to_features(self, table):
return _extract_features(table)