mirror of
https://github.com/EvolutionAPI/adk-python.git
synced 2025-12-23 21:57:44 -06:00
No public description
PiperOrigin-RevId: 748777998
This commit is contained in:
committed by
hangfei
parent
290058eb05
commit
61d4be2d76
@@ -106,9 +106,11 @@ class ResponseEvaluator:
|
||||
eval_dataset = pd.DataFrame(flattened_queries).rename(
|
||||
columns={"query": "prompt", "expected_tool_use": "reference_trajectory"}
|
||||
)
|
||||
eval_task = EvalTask(dataset=eval_dataset, metrics=metrics)
|
||||
|
||||
eval_result = eval_task.evaluate()
|
||||
eval_result = ResponseEvaluator._perform_eval(
|
||||
dataset=eval_dataset, metrics=metrics
|
||||
)
|
||||
|
||||
if print_detailed_results:
|
||||
ResponseEvaluator._print_results(eval_result)
|
||||
return eval_result.summary_metrics
|
||||
@@ -129,6 +131,16 @@ class ResponseEvaluator:
|
||||
metrics.append("rouge_1")
|
||||
return metrics
|
||||
|
||||
@staticmethod
|
||||
def _perform_eval(dataset, metrics):
|
||||
"""This method hides away the call to external service.
|
||||
|
||||
Primarily helps with unit testing.
|
||||
"""
|
||||
eval_task = EvalTask(dataset=dataset, metrics=metrics)
|
||||
|
||||
return eval_task.evaluate()
|
||||
|
||||
@staticmethod
|
||||
def _print_results(eval_result):
|
||||
print("Evaluation Summary Metrics:", eval_result.summary_metrics)
|
||||
|
||||
Reference in New Issue
Block a user