Fix bug where we would convert int "id" groups to float

The problem was that Pandas' iterrows() converted int columns to float. The
solution was using itertuples() instead.

Check pandas-dev/pandas#6509

pipeline/metrics/runner.py

@@ -73,9 +73,11 @@ def main(self, csv_path,
         if groupby:
             votes.insert(0, groupby, votes.index)
 
+        rows = []
+        columns = votes.columns.tolist()
         replace_nan_with_none = lambda df: df.where(pd.notnull(df), None)
-        rows = [collections.OrderedDict(replace_nan_with_none(row))
-                for i, row in votes.iterrows()]
+        for row in replace_nan_with_none(votes).itertuples(False):
+            rows.append(collections.OrderedDict(zip(columns, row)))
         return rows
 
     def calculate_metric(self, votes, metric_method):

pipeline/test/metrics/test_runner.py

@@ -5,6 +5,7 @@
 import io
 import csv
 import collections
+import numpy as np
 
 from pipeline.metrics.runner import Runner
 from pipeline.metrics.runner import Rollcall
@@ -81,6 +82,13 @@ def test_main_groups_correctly_when_not_calculating_metric(self):
                             groupby='party')
         self.assertEqual(res, expected_result)
 
+    def test_main_groups_doesnt_convert_int_groups_to_float(self):
+        csv_path = self._get_csv_path('example_votes_with_metadata.csv')
+
+        res = Runner().main(csv_path, metric_method=None,
+                            groupby='id', name='Joao')
+        self.assertEqual(type(res[0]['id']), np.int64)
+
     def test_remove_unanimous_votes_runs_before_filters(self):
         csv_path = self._get_csv_path('example_votes_with_metadata.csv')
         expected_result = [collections.OrderedDict([