Update final scoring method for BinaryClassifierEfficacy precision & recall metrics (make 0.5 a threshold) (#725)

Author: frances-h

sdmetrics/single_table/data_augmentation/base.py

@@ -227,13 +227,9 @@ def compute_breakdown(
                 f'fixed_{metric_to_fix}_value': trainer.fixed_value,
             },
         }
-        result['score'] = max(
-            0,
-            (
-                result['augmented_data'][f'{cls.metric_name}_score_validation']
-                - result['real_data_baseline'][f'{cls.metric_name}_score_validation']
-            ),
-        )
+        augmented_score = result['augmented_data'][f'{cls.metric_name}_score_validation']
+        baseline_score = result['real_data_baseline'][f'{cls.metric_name}_score_validation']
+        result['score'] = (augmented_score - baseline_score) / 2 + 0.5
         return result
 
     @classmethod

sdmetrics/single_table/data_augmentation/binary_classifier_precision_efficacy.py

@@ -18,8 +18,8 @@ def compute_breakdown(
         metadata,
         prediction_column_name,
         minority_class_label,
-        classifier,
-        fixed_recall_value,
+        classifier='xgboost',
+        fixed_recall_value=0.9,
     ):
         """Compute the score breakdown of the metric."""
         return super().compute_breakdown(
@@ -42,8 +42,8 @@ def compute(
         metadata,
         prediction_column_name,
         minority_class_label,
-        classifier,
-        fixed_recall_value,
+        classifier='xgboost',
+        fixed_recall_value=0.9,
     ):
         """Compute the score of the metric.
 

tests/integration/single_table/data_augmentation/test_binary_classifier_precision_efficacy.py

@@ -162,7 +162,7 @@ def test_with_nan_target_column(self):
                 'classifier': 'XGBoost',
                 'fixed_recall_value': 0.8,
             },
-            'score': 0,
+            'score': 0.48571428571428577,
         }
         assert result_breakdown == expected_result
 
@@ -244,6 +244,6 @@ def test_with_multi_class(self):
                 'classifier': 'XGBoost',
                 'fixed_recall_value': 0.8,
             },
-            'score': 0,
+            'score': 0.4944444444444444,
         }
         assert score_breakdown == expected_score_breakdown

tests/integration/single_table/data_augmentation/test_binary_classifier_recall_efficacy.py

@@ -124,7 +124,7 @@ def test_with_nan_target_column(self):
         )
 
         # Assert
-        assert result_breakdown['score'] in (0, 0.07692307692307698)
+        assert result_breakdown['score'] in (0.5, 0.5384615384615385)
 
     def test_with_minority_being_majority(self):
         """Test the metric when the minority class is the majority class."""
@@ -148,7 +148,7 @@ def test_with_minority_being_majority(self):
         )
 
         # Assert
-        assert score == 0
+        assert score == 0.46153846153846156
 
     def test_with_multi_class(self):
         """Test the metric with multi-class classification.
@@ -175,4 +175,4 @@ def test_with_multi_class(self):
         )
 
         # Assert
-        assert score_breakdown['score'] in (0, 0.07692307692307687)
+        assert score_breakdown['score'] in (0.46153846153846156, 0.5384615384615384)

tests/unit/single_table/data_augmentation/test_base.py

@@ -357,7 +357,7 @@ def test_compute_breakdown(
 
         # Assert
         expected_result = {
-            'score': 0.19999999999999996,
+            'score': 0.6,
             'real_data_baseline': real_data_baseline,
             'augmented_data': augmented_table_result,
             'parameters': {