google.adk.evaluation.local_eval_service.LocalEvalService
https://github.com/google/adk-python/blob/v1.23.0/src/google/adk/evaluation/local_eval_service.py#L112
root_agent
eval_sets_manager
self._metric_evaluator_registry
google.adk.evaluation.metric_evaluator_registry.MetricEvaluatorRegistry
指定がないときは google.adk.evaluation.metric_evaluator_registry.DEFAULT_METRIC_EVALUATOR_REGISTRY
eval_set_results_manager
perform_inference() https://github.com/google/adk-python/blob/v1.23.0/src/google/adk/evaluation/local_eval_service.py#L143
Returns InferenceResult obtained from the Agent as and when they are available.
asyncio.Semaphore使っている
evaluate() https://github.com/google/adk-python/blob/v1.23.0/src/google/adk/evaluation/local_eval_service.py#L192
Returns EvalCaseResult for each item as and when they are available.
_evaluate_single_inference_result()呼び出し
self._eval_set_results_manager.save_eval_set_result()
(google.adk.evaluation.local_eval_set_results_manager.LocalEvalSetResultsManagerであれば.adk/の下に保存)
_evaluate_single_inference_result()
_evaluate_metric_for_eval_case()呼び出し
_evaluate_metric_for_eval_case()
_evaluate_metric()呼び出し
_evaluate_metric() https://github.com/google/adk-python/blob/v1.23.0/src/google/adk/evaluation/local_eval_service.py#L415
metric_evaluator = self._metric_evaluator_registry.get_evaluator(eval_metric=eval_metric)
#google.adk.evaluation.evaluator.Evaluator
metric_evaluator.evaluate_invocations()