scoring.py 文件源码

def benchmark_score_from_local(benchmark_id, training_dir):
    spec = gym.benchmark_spec(benchmark_id)

    directories = []
    for name, _, files in os.walk(training_dir):
        manifests = gym.monitoring.detect_training_manifests(name, files=files)
        if manifests:
            directories.append(name)

    benchmark_results = defaultdict(list)
    for training_dir in directories:
        results = gym.monitoring.load_results(training_dir)

        env_id = results['env_info']['env_id']
        benchmark_result = spec.score_evaluation(env_id, results['data_sources'], results['initial_reset_timestamps'], results['episode_lengths'], results['episode_rewards'], results['episode_types'], results['timestamps'])
        # from pprint import pprint
        # pprint(benchmark_result)
        benchmark_results[env_id].append(benchmark_result)

    return gym.benchmarks.scoring.benchmark_aggregate_score(spec, benchmark_results)