fixed tests

.travis.yml

@@ -9,4 +9,8 @@ script:
   - pytest
 
 notifications:
-  slack: tw-datalab:tySjyrBfi9MdHdiuk5hh8Jbd
+  slack:
+    rooms:
+      - tw-datalab:tySjyrBfi9MdHdiuk5hh8Jbd
+    on_success: change
+    on_failure: always

src/evaluation.py

@@ -7,7 +7,8 @@ def nwrmsle(predictions, targets, weights):
         predictions = np.array([np.nan if x < 0 else x for x in predictions])
     elif type(predictions) == pd.Series:
         predictions[predictions < 0] = np.nan
-    targets[targets < 0] = np.nan
+    targetsf = targets.astype(float)
+    targetsf[targets < 0] = np.nan
     weights = 1 + 0.25 * weights
-    log_square_errors = (np.log(predictions + 1) - np.log(targets + 1)) ** 2
+    log_square_errors = (np.log(predictions + 1) - np.log(targetsf + 1)) ** 2
     return(np.sqrt(np.sum(weights * log_square_errors) / np.sum(weights)))

test/load_and_split_test.py → test/splitter_test.py

@@ -3,11 +3,11 @@
 import pandas as pd
 sys.path.append(os.path.join('..', 'src'))
 sys.path.append(os.path.join('src'))
-import load_and_split
+import splitter
 
 def test_get_validation_period():
     latest_date = pd.to_datetime('2017-11-22')
-    actual_begin_date, actual_end_date = load_and_split.get_validation_period(latest_date)
+    actual_begin_date, actual_end_date = splitter.get_validation_period(latest_date)
     expected_begin_date = pd.to_datetime('2017-11-01')
     expected_end_date = pd.to_datetime('2017-11-16')
     assert actual_begin_date == expected_begin_date
@@ -22,6 +22,6 @@ def test_split_validation_train_by_validation_period():
     validation_end_date = pd.to_datetime('2017-11-30')
     d = {'date': [date1, date2, date3, date4], 'col2': [3, 4, 5, 6]}
     df = pd.DataFrame(data=d)
-    df_train, df_validation = load_and_split.split_validation_train_by_validation_period(df, validation_begin_date, validation_end_date)
+    df_train, df_validation = splitter.split_validation_train_by_validation_period(df, validation_begin_date, validation_end_date)
     assert df_train.shape[0] == 1
     assert df_validation.shape[0] == 2