`trw.callbacks.callback_reporting_layer_statistics`¶

Module Contents¶

Classes¶

CallbackReportingLayerStatistics

Report the activation and gradient statistics layer by layer

Functions¶

`generic_tracing`()	Trace only basic building blocks to avoid too much clutter
`collect_gradient`(model, gradient_store)	Collect the gradient of each parameter of a given model
`aggregate_stats`(all_stats, batch_stat)
`aggregate_stats_end`(all_stats)
`calculate_stats_gradient`(model, sequence, nb_samples, aggregate_stats_fn=aggregate_stats, aggregate_stats_end_fn=aggregate_stats_end, modules_type_to_trace=generic_tracing())	Collect the activation statistics and the gradient update stats for each layer

Attributes¶

logger

trw.callbacks.callback_reporting_layer_statistics.logger¶

trw.callbacks.callback_reporting_layer_statistics.generic_tracing()¶: Trace only basic building blocks to avoid too much clutter

trw.callbacks.callback_reporting_layer_statistics.collect_gradient(model, gradient_store)¶

Collect the gradient of each parameter of a given model :param model: the model :param gradient_store: where to store the parameter gradients

Returns:

trw.callbacks.callback_reporting_layer_statistics.aggregate_stats(all_stats, batch_stat)¶

trw.callbacks.callback_reporting_layer_statistics.aggregate_stats_end(all_stats)¶

trw.callbacks.callback_reporting_layer_statistics.calculate_stats_gradient(model, sequence, nb_samples, aggregate_stats_fn=aggregate_stats, aggregate_stats_end_fn=aggregate_stats_end, modules_type_to_trace=generic_tracing())¶

Collect the activation statistics and the gradient update stats for each layer

Returns: a tuple (gradient stats, activation stats)

class trw.callbacks.callback_reporting_layer_statistics.CallbackReportingLayerStatistics(dataset_name=None, split_name=None, nb_samples=500, table_name='layer')¶

Bases: trw.callbacks.callback.Callback