"""Global metrics management for training and evaluation.
This module provides a simple global registry for metrics that can be accessed
and updated throughout the training process. It maintains a global dictionary
of metrics that can be shared across different components of the framework.
"""
GLOBAL_METRICS = {}
MOS_METRICS = {}
def get_log_metrics():
"""Get the global metrics dictionary.
Returns:
dict: The global metrics dictionary containing all registered metrics.
Example:
>>> metrics = get_log_metrics()
>>> print(metrics)
{'accuracy': 0.95, 'loss': 0.05}
"""
return GLOBAL_METRICS
def get_mos_metrics():
"""Get the metrics dictionary for MOS.
Returns:
dict: The global mos metrics dictionary containing all registered metrics.
Example:
>>> metrics = get_mos_metrics()
>>> print(metrics)
{'accuracy': 0.95, 'loss': 0.05}
"""
return MOS_METRICS
[docs]
def add_metric(name, metric, report_to_mos=True):
"""Add or update a metric in the global metrics registry.
Args:
name (str): The name of the metric to add or update.
metric: The metric value to store. Can be any type (float, int, tensor, etc.).
report_to_mos (bool, optional): Whether to report the metric to Mos. Defaults to False.
Example:
>>> add_metric("accuracy", 0.95)
>>> add_metric("loss", 0.05)
>>> add_metric("learning_rate", 0.001)
"""
global GLOBAL_METRICS, MOS_METRICS
GLOBAL_METRICS[name] = metric
if report_to_mos:
MOS_METRICS[name] = metric