diff --git a/atroposlib/envs/base.py b/atroposlib/envs/base.py index 6ee5f2b8..b94f0995 100644 --- a/atroposlib/envs/base.py +++ b/atroposlib/envs/base.py @@ -721,7 +721,7 @@ class BaseEnv(ABC): eval_result = { "config_general": { "model_name": model_name, - "total_evaluation_time_secondes": str(end_time - start_time), + "total_evaluation_time_seconds": str(end_time - start_time), "generation_parameters": merged_gen_params, }, "results": { diff --git a/atroposlib/type_definitions.py b/atroposlib/type_definitions.py index bf3ffd3c..b435f677 100644 --- a/atroposlib/type_definitions.py +++ b/atroposlib/type_definitions.py @@ -74,7 +74,7 @@ GameHistory = List[GameStep] class EvaluationConfigGeneral(TypedDict): """Configuration section of evaluation results.""" - total_evaluation_time_secondes: str + total_evaluation_time_seconds: str model_name: Optional[str] generation_parameters: Dict[str, Any] diff --git a/atroposlib/utils/metrics.py b/atroposlib/utils/metrics.py index 0a1c75d4..06066e21 100644 --- a/atroposlib/utils/metrics.py +++ b/atroposlib/utils/metrics.py @@ -4,13 +4,15 @@ import numpy as np def get_std_min_max_avg(name: str, data: list, metrics_dict: dict) -> dict: """ Calculate the standard deviation, minimum, maximum, and average of a list of numbers. - Adds it to the wandb dict for logging. + Adds it to the metrics dict for logging. Args: - data (list): A list of numbers. + name: The base name for the metrics keys. + data: A list of numbers to compute statistics from. + metrics_dict: Dictionary to add the computed metrics to. Returns: - dict: A dictionary containing the standard deviation, minimum, maximum, and average. + The updated metrics dictionary with added statistics (mean, std, max, min). """ metrics_dict[f"{name}_mean"] = np.mean(data) metrics_dict[f"{name}_std"] = np.std(data)