Runner di set di dati Batch

I BatchEvaluationRunner delegati si occupano interamente della raccolta e della valutazione fino al servizio tramite le API and. StartBatchEvaluation GetBatchEvaluation Dopo aver richiamato l'agente per ogni scenario, il runner invia un processo in batch ed esegue i sondaggi fino al completamento, restituendo risultati aggregati.

Usa il batch runner quando hai bisogno di punteggi aggregati per più sessioni senza dover gestire personalmente la raccolta degli intervalli; per misurazioni di base, set di dati di grandi dimensioni e confronti. pre/post

Come funziona

Il runner elabora gli scenari in quattro fasi:

Agente invoker

Il runner richiede un agent invoker, un callable che richiama il tuo agente per un solo turno. L'invoker è indipendente dal framework: puoi chiamare il tuo agente tramite invoke_agent_runtime boto3, una chiamata di funzione diretta, una richiesta HTTP o qualsiasi altro metodo.

import json
import boto3
from bedrock_agentcore.evaluation import AgentInvokerInput, AgentInvokerOutput

REGION       = "<region-code>"
AGENT_ARN    = "arn:aws:bedrock-agentcore:<region-code>:<account-id>:runtime/<agent-id>"
LOG_GROUP    = "/aws/bedrock-agentcore/runtimes/<agent-id>-DEFAULT"
SERVICE_NAME = "<agent-id>.DEFAULT"

agentcore_client = boto3.client("bedrock-agentcore", region_name=REGION)

def agent_invoker(invoker_input: AgentInvokerInput) -> AgentInvokerOutput:
    payload = invoker_input.payload
    if isinstance(payload, str):
        payload = json.dumps({"prompt": payload}).encode()
    elif isinstance(payload, dict):
        payload = json.dumps(payload).encode()

    print(f"[{invoker_input.session_id}] > sending payload: {payload.decode()}")
    response = agentcore_client.invoke_agent_runtime(
        agentRuntimeArn=AGENT_ARN,
        runtimeSessionId=invoker_input.session_id,
        payload=payload,
    )
    response_body = response["response"].read()
    print(f"[{invoker_input.session_id}] < received response: {response_body.decode()}")
    return AgentInvokerOutput(agent_output=json.loads(response_body))

Esempio

L'esempio seguente carica un set di dati da un file JSON ed esegue la valutazione in batch. Per il formato del set di dati, vedere Schema del set di dati.

from bedrock_agentcore.evaluation import (
    BatchEvaluationRunner,
    BatchEvaluationRunConfig,
    BatchEvaluatorConfig,
    CloudWatchDataSourceConfig,
    FileDatasetProvider,
)

# Load dataset from a local file (see Dataset schema for format)
dataset = FileDatasetProvider("dataset.json").get_dataset()

# Or load from the Dataset Management service
from bedrock_agentcore.evaluation import DatasetClient, DatasetManagementServiceProvider
ds_client = DatasetClient(region_name=REGION)
dataset = DatasetManagementServiceProvider(dataset_id="my-dataset-id", client=ds_client).get_dataset()

# Configure the batch evaluation
config = BatchEvaluationRunConfig(
    batch_evaluation_name="dataset-batch-eval",
    evaluator_config=BatchEvaluatorConfig(
        evaluator_ids=[
            "Builtin.GoalSuccessRate",
            "Builtin.Correctness",
            "Builtin.TrajectoryExactOrderMatch",
            "Builtin.Helpfulness",
        ],
    ),
    data_source=CloudWatchDataSourceConfig(
        service_names=[SERVICE_NAME],
        log_group_names=[LOG_GROUP],
        ingestion_delay_seconds=180,
    ),
    polling_timeout_seconds=1800,
    polling_interval_seconds=30,
)

# Run
runner = BatchEvaluationRunner(region=REGION)
result = runner.run_dataset_evaluation(
    agent_invoker=agent_invoker,
    dataset=dataset,
    config=config,
)

# Display aggregate results
print(f"Status: {result.status}")
print(f"Batch evaluation ID: {result.batch_evaluation_id}")

if result.evaluation_results:
    er = result.evaluation_results
    print(f"Sessions completed: {er.number_of_sessions_completed}")
    print(f"Sessions failed:    {er.number_of_sessions_failed}")
    print(f"Total sessions:     {er.total_number_of_sessions}")

    for summary in er.evaluator_summaries or []:
        avg = summary.statistics.average_score if summary.statistics else None
        print(f"  {summary.evaluator_id}: avg={avg}")

Recupero dei dettagli per sessione

I risultati aggregati mostrano le medie di tutte le sessioni. Per visualizzare i punteggi per sessione e per valutatore, recupera gli eventi di valutazione da: CloudWatch

if result.output_data_config:
    events = runner.fetch_evaluation_events(result)
    print(f"\nEvaluation events: {len(events)}")
    for ev in events:
        attrs = ev.get("attributes", {})
        print(f"  session: {attrs.get('session.id', '')[:40]}")
        print(f"  evaluator: {attrs.get('gen_ai.evaluation.name')}")
        print(f"  score: {attrs.get('gen_ai.evaluation.score.value')}")
        print(f"  label: {attrs.get('gen_ai.evaluation.score.label')}")
        print()

Informazioni di riferimento sulla configurazione

BatchEvaluationRunConfig(
    batch_evaluation_name="my-batch-eval",   # Job name
    evaluator_config=BatchEvaluatorConfig(
        evaluator_ids=["Builtin.GoalSuccessRate"],
    ),
    data_source=CloudWatchDataSourceConfig(
        service_names=["MyAgent.DEFAULT"],   # Exactly 1 service name
        log_group_names=[LOG_GROUP],         # 1-5 log group names
        ingestion_delay_seconds=180,         # Wait for CW ingestion (default: 180)
    ),
    polling_timeout_seconds=1800,            # Max wait for job completion (default: 1800)
    polling_interval_seconds=30,             # Poll interval (default: 30)
    simulation_config=None,                  # Set SimulationConfig for simulated scenarios
)

Struttura dei risultati

Il corridore restituisce unBatchEvaluationResult:

BatchEvaluationResult
  ├── batch_evaluation_id: str
  ├── batch_evaluation_arn: str
  ├── batch_evaluation_name: str
  ├── status: str
  ├── created_at: datetime
  ├── evaluation_results: Optional[BatchEvaluationSummary]
  │     ├── number_of_sessions_completed: int
  │     ├── number_of_sessions_in_progress: int
  │     ├── number_of_sessions_failed: int
  │     ├── number_of_sessions_ignored: int
  │     ├── total_number_of_sessions: int
  │     └── evaluator_summaries: List
  │           ├── evaluator_id: str
  │           ├── statistics.average_score: float
  │           ├── total_evaluated: int
  │           └── total_failed: int
  ├── error_details: Optional[List[str]]
  ├── agent_invocation_failures: List[FailedScenario]
  └── output_data_config: Optional[CloudWatchOutputDataConfig]
        ├── log_group_name: str
        └── log_stream_name: str

Gestione degli errori