super_gradients.training.legacy package

Submodules

super_gradients.training.legacy.utils module

super_gradients.training.legacy.utils.prefetch_dataset(dataset, num_workers=4, batch_size=32, device=None, half=False)[source]
class super_gradients.training.legacy.utils.PrefetchDataLoader(dataloader, device, half=False)[source]

Bases: object

async_prefech()[source]
super_gradients.training.legacy.utils.init_params(net)[source]

Init layer parameters.

super_gradients.training.legacy.utils.format_time(seconds)[source]
super_gradients.training.legacy.utils.is_better(new_metric, current_best_metric, metric_to_watch='acc')[source]

Determines which of the two metrics is better, the higher if watching acc or lower when watching loss :param new_metric: the new metric :param current_best_metric: the compared to metric :param metric_to_watch: acc or loss :return: bool, True if new metric is better than current

super_gradients.training.legacy.utils.makedirs_if_not_exists(dir_path: str)[source]
make new directory in dir_path if it doesn’t exists

:param dir_path - full path of directory

Module contents