[LCORE-648] Fix processing of float('NaN') values when OutputParserException

src/lightspeed_evaluation/core/metrics/ragas.py

@@ -1,5 +1,6 @@
 """Ragas metrics evaluation using LLM Manager."""
 
+import math
 from typing import Any, Optional
 
 from datasets import Dataset
@@ -90,7 +91,7 @@ def evaluate(
             return None, f"Unsupported Ragas metric: {metric_name}"
 
         try:
-            return self.supported_metrics[metric_name](
+            result = self.supported_metrics[metric_name](
                 conv_data, scope.turn_idx, scope.turn_data, scope.is_conversation
             )
         except BrokenPipeError as e:
@@ -100,16 +101,27 @@ def evaluate(
                 f"connection issue: {str(e)}",
             )
         except OSError as e:
+            err_msg = f"Ragas {metric_name} evaluation failed: {str(e)}"
             if e.errno == 32:  # Broken pipe
-                return (
-                    None,
+                err_msg = (
                     f"Ragas {metric_name} evaluation failed due to broken pipe "
-                    f"(network/LLM timeout): {str(e)}",
+                    f"(network/LLM timeout): {str(e)}"
                 )
-            return None, f"Ragas {metric_name} evaluation failed: {str(e)}"
+            return None, err_msg
         except (RuntimeError, ValueError, TypeError, ImportError) as e:
             return None, f"Ragas {metric_name} evaluation failed: {str(e)}"
 
+        # Ragas returns float('NaN') when it cannot parse the output from the
+        # LLM (OutputParserException)
+        if result[0] is not None and math.isnan(result[0]):
+            return (
+                None,
+                f"Ragas {metric_name} evaluation failed due to malformed "
+                "output from the LLM",
+            )
+
+        return result
+
     def _evaluate_response_relevancy(
         self,
         _conv_data: Any,