core.common.hpo_utils

Copyright (c) Meta, Inc. and its affiliates.

This source code is licensed under the MIT license found in the LICENSE file in the root directory of this source tree.

Functions

tune_reporter(→ None)	Wrapper function for tune.report()
label_metric_dict(metric_dict, split)

Module Contents

core.common.hpo_utils.tune_reporter(iters, train_metrics, val_metrics, test_metrics=None, metric_to_opt: str = 'val_loss', min_max: str = 'min') → None

Wrapper function for tune.report()

Parameters:

• iters (dict) – dict with training iteration info (e.g. steps, epochs)

• train_metrics (dict) – train metrics dict

• val_metrics (dict) – val metrics dict

• test_metrics (dict, optional) – test metrics dict, default is None

• metric_to_opt (str, optional) – str for val metric to optimize, default is val_loss

• min_max (str, optional) – either “min” or “max”, determines whether metric_to_opt is to be minimized or maximized, default is min

core.common.hpo_utils.label_metric_dict(metric_dict, split)