Merge pull request #3 from futurice/feature-math-coding

update outputs for eval dags

math_coding/experiment.yaml

@@ -4,7 +4,7 @@ flow: flows/math_standard_flow
 connections:
 - name: aoai
   connection_type: AzureOpenAIConnection
-  api_base: https://demoopenaiexamples.openai.azure.com/
+  api_base: https://edge-10x-ai-services.cognitiveservices.azure.com
   api_version: 2023-07-01-preview
   api_key: ${api_key}
   api_type: azure

math_coding/flows/math_evaluation_flow/flow.dag.yaml

@@ -8,8 +8,11 @@ inputs:
     default: "2"
 outputs:
   score:
-    type: string
+    type: integer
     reference: ${line_process.output}
+  metrics:
+    type: object
+    reference: ${aggregate.output}
 nodes:
 - name: line_process
   type: python

math_coding/flows/math_evaluation_flow/line_process.py

@@ -2,24 +2,23 @@
 
 
 @tool
-def line_process(groundtruth: str, prediction: str) -> int:
-
+def line_process(groundtruth: str, prediction: str):
     processed_result = 0
 
     if prediction == "JSONDecodeError" or prediction.startswith("Unknown Error:"):
         processed_result = -1
-        return processed_result
+        return {"output": processed_result}
 
     try:
         groundtruth = float(groundtruth)
         prediction = float(prediction)
     except ValueError:
         processed_result = -1
-        return processed_result
+        return {"output": processed_result}
 
     if round(prediction, 2) == round(groundtruth, 2):
         processed_result = 1
-    return processed_result
+    return {"output": processed_result}
 
 
 if __name__ == "__main__":