trw.callbacks.callback_save_last_model

Module Contents

Classes

ModelWithLowestMetricBase

ModelWithLowestMetric

CallbackSaveLastModel

Save the current model to disk as well as metadata (history, outputs, infos).

Functions

exclude_large_embeddings(outputs: trw.basic_typing.Datasets, counts_greater_than=10000) → Optional[trw.basic_typing.Datasets]

Remove from the outputs embeddings larger than a specified threshold.

should_not_export_model(last_step, revert_if_nan_metrics)

Attributes

logger

trw.callbacks.callback_save_last_model.logger
class trw.callbacks.callback_save_last_model.ModelWithLowestMetricBase
abstract update(self, metric_value, model, metadata, root_path)
class trw.callbacks.callback_save_last_model.ModelWithLowestMetric(dataset_name, split_name, output_name, metric_name, minimum_metric=0.2)

Bases: ModelWithLowestMetricBase

update(self, metric_value, model, metadata, root_path)

Check the metrics and export the model if thresholds are satisfied

trw.callbacks.callback_save_last_model.exclude_large_embeddings(outputs: trw.basic_typing.Datasets, counts_greater_than=10000) → Optional[trw.basic_typing.Datasets]

Remove from the outputs embeddings larger than a specified threshold.

Parameters

  • outputs – the outputs to check

  • counts_greater_than – the number of elements above which the embedding will be stripped

Returns

outputs

trw.callbacks.callback_save_last_model.should_not_export_model(last_step, revert_if_nan_metrics)
class trw.callbacks.callback_save_last_model.CallbackSaveLastModel(model_name='last', with_outputs=False, is_versioned=False, rolling_size=None, keep_model_with_best_metric: ModelWithLowestMetric = None, revert_if_nan_metrics: Optional[Sequence[str]] = ('loss',), post_process_outputs: Optional[Callable[[trw.basic_typing.Datasets], trw.basic_typing.Datasets]] = exclude_large_embeddings)

Bases: trw.callbacks.callback.Callback

Save the current model to disk as well as metadata (history, outputs, infos).

This callback can be used during training (e.g., checkpoint) or at the end of the training.

Optionally, record the best model for a given dataset, split, output and metric.

__call__(self, options, history, model, losses, outputs, datasets, datasets_infos, callbacks_per_batch, **kwargs)