Correctly handle multi-channel data and labels

Author: acrellin

cesium_app/handlers/feature.py

@@ -131,25 +131,30 @@ class PrecomputedFeaturesHandler(BaseHandler):
     @auth_or_token
     def post(self):
         data = self.get_json()
-        print(data)
         if data['datasetID'] not in [None, 'None']:
             dataset = Dataset.query.filter(Dataset.id == data['datasetID']).one()
         else:
             dataset = None
         current_project = Project.get_if_owned_by(data['projectID'],
                                                   self.current_user)
         feature_data = StringIO(data['dataFile']['body'])
-        fset = pd.read_csv(feature_data, sep=',', index_col='ts_name')
-        try:
-            labels = fset.pop('labels').values
-        except KeyError:
+        fset = pd.read_csv(feature_data, index_col=0, header=[0, 1])
+        if 'labels' in fset:
+            labels = fset.pop('labels').values.ravel()
+            if labels.dtype == 'O':
+                labels = [str(label) for label in labels]
+        else:
             labels = [None]
         fset_path = pjoin(
             self.cfg['paths:features_folder'],
             '{}_{}.npz'.format(uuid.uuid4(), data['dataFile']['name']))
+
         featurize.save_featureset(fset, fset_path, labels=labels)
 
-        features_list = fset.columns.tolist()
+        # Meta-features will have channel values of an empty string or a string
+        # beginning with 'Unnamed:'
+        features_list = [el[0] for el in fset.columns.tolist() if
+                         (el[1] != '' and not el[1].startswith('Unnamed:'))]
 
         featureset = Featureset(name=data['featuresetName'],
                                 file_uri=fset_path,