Update bias_auc.py

bias_auc.py

@@ -120,25 +120,25 @@ class BiasAUC(evaluate.Metric):
             # subgroup positive
             elif subgroup in target and label == target_class:
                  yield label, result[target_class]
-            
+
+    def _get_auc_score(self, gen_func, *args, **kwargs):
+        try:
+            y_trues, y_preds = zip(*self.gen_func(args))
+            score = roc_auc_score(y_trues, y_preds)
+        except ValueError:
+            print(f"Sample not sufficient for target clases '{args[-1]}' subgroup metric (need correct and incorrect predictions for '{args[-1]}')")
+            score = np.nan
+        return score
+  
     def _auc_by_group(self, target, label, output, subgroup):
         """ Compute bias AUC metrics 
         """
-    
-        y_trues, y_preds = zip(*self._genreate_subgroup(target, label, output, subgroup))
-        subgroup_auc_score = roc_auc_score(y_trues, y_preds)
-        
-        y_trues, y_preds = zip(*self._genreate_bpsn(target, label, output, subgroup))
-        bpsn_auc_score = roc_auc_score(y_trues, y_preds)
-
-        y_trues, y_preds = zip(*self._genreate_bnsp(target, label, output, subgroup))
-        bnsp_auc_score = roc_auc_score(y_trues, y_preds)
+        return {
+            'Subgroup' : self._get_auc_score(self._genreate_subgroup, target, label, output, subgroup),
+            'BPSN' : self._get_auc_score(self._genreate_bpsn, target, label, output, subgroup),
+            'BNSP' : self._get_auc_score(self._genreate_bnsp, target, label, output, subgroup)
+        }
 
-
-        return {'Subgroup' : subgroup_auc_score,
-                'BPSN' : bpsn_auc_score,
-                'BNSP' : bnsp_auc_score}
-    
     def _update_overall(self, result, labels, outputs, power_value=-5):
         """Compute the generalized mean of Bias AUCs"""
         result['Overall'] = {}
@@ -149,9 +149,11 @@ class BiasAUC(evaluate.Metric):
             mean_value = np.power(np.sum(metric_values)/(len(result) - 1),  1/power_value)
             result['Overall'][f"{metric} generalized mean"] = mean_value
         y_preds = [output[1] for output in outputs]
-        result['Overall']["Overall AUC"] = roc_auc_score(labels, y_preds)
-        return result
-            
+        try:
+            result['Overall']["Overall AUC"] = roc_auc_score(labels, y_preds)
+        except ValueError:
+            result['Overall']["Overall AUC"] = np.nan
+        return result        
     
     def _compute(self, target, label, output, subgroups=None):
         if subgroups is None: