[fix] avoid failure when partial success of some raiservice evaluators (#44429)

Co-authored-by: zyysurely <[email protected]>

Author: Zyysurely

sdk/evaluation/azure-ai-evaluation/azure/ai/evaluation/_evaluate/_evaluate.py

@@ -235,6 +235,10 @@ def _aggregate_label_defect_metrics(df: pd.DataFrame) -> Tuple[List[str], Dict[s
 
 
 def _process_rows(row, detail_defect_rates):
+    # Ignore the failed rows (NaN, None, etc.), which is not a expected dict
+    if _is_none_or_nan(row):
+        return detail_defect_rates
+
     for key, value in row.items():
         if key not in detail_defect_rates:
             detail_defect_rates[key] = []

sdk/evaluation/azure-ai-evaluation/tests/unittests/test_evaluate.py

@@ -37,6 +37,8 @@
     _apply_target_to_data,
     _rename_columns_conditionally,
     _convert_results_to_aoai_evaluation_results,
+    _process_rows,
+    _aggregate_label_defect_metrics,
 )
 from azure.ai.evaluation._evaluate._utils import _convert_name_map_into_property_entries
 from azure.ai.evaluation._evaluate._utils import _apply_column_mapping, _trace_destination_from_project_scope
@@ -719,6 +721,31 @@ def test_general_aggregation(self):
         assert "bad_thing.boolean_with_nan" not in aggregation
         assert "bad_thing.boolean_with_none" not in aggregation
 
+    def test_aggregate_label_defect_metrics_with_nan_in_details(self):
+        """Test that NaN/None values in details column are properly ignored during aggregation."""
+        data = {
+            "evaluator.protected_material_label": [True, False, True, False],
+            "evaluator.protected_material_details": [
+                {"detail1": 1, "detail2": 0},
+                np.nan,  # Failed evaluation
+                {"detail1": 0, "detail2": 1},
+                None,  # Another failure case
+            ],
+        }
+        df = pd.DataFrame(data)
+
+        label_cols, defect_rates = _aggregate_label_defect_metrics(df)
+
+        # Should calculate defect rate for label column (all 4 rows)
+        assert "evaluator.protected_material_defect_rate" in defect_rates
+        assert defect_rates["evaluator.protected_material_defect_rate"] == 0.5
+
+        # Should calculate defect rates for detail keys (only from 2 valid dict rows)
+        assert "evaluator.protected_material_details.detail1_defect_rate" in defect_rates
+        assert "evaluator.protected_material_details.detail2_defect_rate" in defect_rates
+        assert defect_rates["evaluator.protected_material_details.detail1_defect_rate"] == 0.5
+        assert defect_rates["evaluator.protected_material_details.detail2_defect_rate"] == 0.5
+
     @pytest.mark.skip(reason="Breaking CI by crashing pytest somehow")
     def test_optional_inputs_with_data(self, questions_file, questions_answers_basic_file):
         from test_evaluators.test_inputs_evaluators import HalfOptionalEval, NoInputEval, NonOptionalEval, OptionalEval