bugfix auc (#7)

CHANGELOG.txt

@@ -2,6 +2,12 @@
 CHANGELOG
 =========
 
+-------------------------------------------------------------------------------
+July 12, 2021 1.2.2
+-------------------------------------------------------------------------------
+
+- Fixed bug with Area under the Curve (AUC) metric when only one class exists.
+
 -------------------------------------------------------------------------------
 June 23, 2021 1.2.1
 -------------------------------------------------------------------------------

jurity/_version.py

@@ -2,4 +2,4 @@
 # Copyright FMR LLC <opensource@fidelity.com>
 # SPDX-License-Identifier: Apache-2.0
 
-__version__ = "1.2.1"
+__version__ = "1.2.2"

jurity/recommenders/auc.py

@@ -142,7 +142,10 @@ def get_score(self, actual_results: pd.DataFrame, predicted_results: pd.DataFram
     @staticmethod
     def _get_results(results: List[np.ndarray]) -> float:
         results = np.concatenate(results)
-        return roc_auc_score(results[:, 0], results[:, 1])
+        if len(set(results[:, 0])) > 1:
+            return roc_auc_score(results[:, 0], results[:, 1])
+        else:
+            return np.nan
 
     def _get_extended_results(self, results: List[np.ndarray]) -> dict:
         auc = self._get_results(results)

tests/test_reco_binary.py

@@ -4,6 +4,7 @@
 
 import unittest
 
+import numpy as np
 import pandas as pd
 
 from jurity.recommenders import BinaryRecoMetrics
@@ -106,3 +107,17 @@ def test_auc_extended(self):
 
         self.assertEqual(0.78125, results['auc'])
         self.assertEqual(12, results['support'])
+
+    def test_auc_one_class(self):
+        # Test immediate calculation of AUC
+        metric = BinaryRecoMetrics.AUC(click_column='click')
+        actual = pd.DataFrame({Constants.user_id: [1, 2, 3, 4],
+                               Constants.item_id: [1, 2, 0, 3],
+                               'click': [0, 0, 0, 0]})
+
+        predicted = pd.DataFrame({Constants.user_id: [1, 2, 3, 4],
+                                  Constants.item_id: [1, 2, 2, 3],
+                                  'click': [0.1, 0.9, 0.1, 0.1]})
+
+        auc = metric.get_score(actual, predicted)
+        self.assertTrue(np.isnan(auc))