sparseml.keras.utils package

Submodules

sparseml.keras.utils.callbacks module

Built-in callbacks for Keras

class sparseml.keras.utils.callbacks.LoggerSettingCallback(loggers: Union[sparseml.keras.utils.logger.KerasLogger, List[sparseml.keras.utils.logger.KerasLogger]])

Bases: tensorflow.python.keras.callbacks.Callback

Class to help correctly set logging modes for callbacks that rely on KerasLogger. All callbacks using KerasLogger should derive from this class.

Parameters

loggers – logger or list of loggers

on_epoch_begin(epoch, logs=None)

Called at the begin of a training epoch

Parameters

• epoch – epoch index

• logs – dictionary of logs (see Keras Callback doc)

on_epoch_end(epoch, logs=None)

Called at the end of a training epoch

Parameters

• epoch – epoch index

• logs – dictionary of logs (see Keras Callback doc)

on_predict_batch_begin(batch, logs=None)

Called at the begin of a batch in prediction

Parameters

• batch – batch index in current epoch

• logs – dictionary of logs (see Keras Callback doc)

on_predict_batch_end(batch, logs=None)

Called at the end of a batch in prediction

Parameters

• batch – batch index in current epoch

• logs – dictionary of logs (see Keras Callback doc)

on_predict_begin(logs=None)

Called at the begin of prediction

Parameters

logs – dictionary of logs (see Keras Callback doc)

on_predict_end(logs=None)

Called at the end of prediction

Parameters

logs – dictionary of logs (see Keras Callback doc)

on_test_batch_begin(batch, logs=None)

Called at the begin of a batch in evaluation

Parameters

• batch – batch index in current epoch

• logs – dictionary of logs (see Keras Callback doc)

on_test_batch_end(batch, logs=None)

Called at the end of a batch in evaluation

Parameters

• batch – batch index in current epoch

• logs – dictionary of logs (see Keras Callback doc)

on_test_begin(logs=None)

Called at the begin of evaluation

Parameters

logs – dictionary of logs (see Keras Callback doc)

on_test_end(logs=None)

Called at the end of evaluation

Parameters

logs – dictionary of logs (see Keras Callback doc)

on_train_batch_begin(batch, logs=None)

Called at the begin of a batch in training

Parameters

• batch – batch index in current epoch

• logs – dictionary of logs (see Keras Callback doc)

on_train_batch_end(batch, logs=None)

Called at the end of a batch in training

Parameters

• batch – batch index in current epoch

• logs – dictionary of logs (see Keras Callback doc)

on_train_begin(logs=None)

Called at the begin of training

Parameters

logs – dictionary of logs (see Keras Callback doc)

on_train_end(logs=None)

Called at the end of training

Parameters

logs – dictionary of logs (see Keras Callback doc)

class sparseml.keras.utils.callbacks.LossesAndMetricsLoggingCallback(loggers: Union[sparseml.keras.utils.logger.KerasLogger, List[sparseml.keras.utils.logger.KerasLogger]], start_step: Union[tensorflow.python.framework.ops.Tensor, int] = 0)

Bases: sparseml.keras.utils.callbacks.LoggerSettingCallback

Callback to log all losses and metrics

Parameters

• loggers – logger or list of loggers

• start_step – a start step tensor when this callback starts to take effect

on_epoch_end(epoch, logs=None)

Called at the end of a training epoch

Parameters

• epoch – epoch index

• logs – dictionary of logs (see Keras Callback doc)

on_test_end(logs=None)

Called at the end of evaluation

Parameters

logs – dictionary of logs (see Keras Callback doc)

on_train_batch_end(batch, logs=None)

Called at the end of a batch in training

Parameters

• batch – batch index in current epoch

• logs – dictionary of logs (see Keras Callback doc)

on_train_begin(logs=None)

Called at the begin of training

Parameters

logs – dictionary of logs (see Keras Callback doc)

sparseml.keras.utils.compat module

sparseml.keras.utils.compat.assign(lhs, rhs, name=None)

sparseml.keras.utils.exporter module

Export Keras models to the local device.

class sparseml.keras.utils.exporter.ModelExporter(model: tensorflow.python.keras.engine.training.Model, output_dir: str)

Bases: object

An exporter for exporting Keras models into ONNX format as well as numpy arrays for the input and output tensors.

Parameters

• model – the module to export

• output_dir – the directory to export the module and extras to

export_h5(name: str = 'model.h5')

Export the Keras model as a single HDF5 file in the output_dir/keras directory

Parameters

name – name to export file with. default is model.h5

export_keras()

Export the model Keras files to the output_dir/keras directory

export_onnx(name: str = 'model.onnx', opset: int = 11, doc_string: str = '', debug_mode: bool = True, raise_on_tf_support: bool = True, **kwargs)

Export an ONNX file for the current model.

Parameters

• name – name of the onnx file to save

• opset – onnx opset to use for exported model. Default is 11

• doc_string – optional doc string for exported ONNX model

• debug_mode – debug mode, default to True, passed into convert_keras

• kwargs – additional parameters passed into convert_keras

export_samples(sample_batches: List[Any], sample_labels: Optional[List[Any]] = None, exp_counter: int = 0)

Export a set list of sample batches as inputs and outputs through the model.

Parameters

• sample_batches – a list of the sample batches to feed through the module for saving inputs and outputs

• sample_labels – an optional list of sample labels that correspond to the the batches for saving

• exp_counter – the counter to start exporting the tensor files at

sparseml.keras.utils.logger module

Contains code for loggers that help visualize the information from each modifier

class sparseml.keras.utils.logger.KerasLogger(name: str, update_freq: Union[str, int] = 'epoch', **kwargs)

Bases: abc.ABC

Base class that all modifier loggers shall implement.

Parameters

• name – name given to the logger, used for identification

• update_freq – define when logging should happen - “epoch”: log at the end of each epoch - “batch”: log at the end of each training batch - an integer value: the number of batches before the next logging should be

abstract log_scalar(tag: str, value: float, step: Optional[int] = None, **kwargs)

Parameters

• tag – identifying tag to log the value with

• value – value to save

• step – global step for when the value was taken

property mode

Mode the current logger is at a current time

Returns

a LoggingMode

property name

name given to the logger, used for identification

Type

return

property update_freq

class sparseml.keras.utils.logger.LoggingMode(value)

Bases: enum.Enum

Some logger changes its logging behavior (e.g. the destination it logs to) based on whether it’s running in train, validation or predict mode

This enum defines the mode a logger could be in during its lifetime

PREDICT = 'predict'

TEST = 'validation'

TRAIN = 'train'

class sparseml.keras.utils.logger.PythonLogger(name: str = 'python', update_freq: Union[str, int] = 'epoch', logger: Optional[logging.Logger] = None)

Bases: sparseml.keras.utils.logger.KerasLogger

Modifier logger that handles printing values into a python logger instance.

Parameters

• name – name given to the logger, used for identification;

• update_freq – update frequency (see attribute info in the base class)

• logger – a logger instance to log to, if None then will create it’s own defaults to python

log_scalar(tag: str, value: float, step: Optional[int] = None, **kwargs)

Parameters

• tag – identifying tag to log the value with

• value – value to save

• step – global step for when the value was taken

• wall_time – global wall time for when the value was taken, defaults to time.time()

class sparseml.keras.utils.logger.TensorBoardLogger(name: str = 'tensorboard', update_freq: Union[str, int] = 'epoch', log_dir: str = 'logs')

Bases: sparseml.keras.utils.logger.KerasLogger

Modifier logger that handles outputting values into a TensorBoard log directory for viewing in TensorBoard.

Parameters

• name – name given to the logger, used for identification; defaults to tensorboard

• update_freq – update frequency (see attribute info in the base class)

• log_dir – the path to create a SummaryWriter at. writer must be None to use if not supplied (and writer is None), will create a TensorBoard dir in cwd

log_scalar(tag: str, value: float, step: Optional[int] = None, **kwargs)

Parameters

• tag – identifying tag to log the value with

• value – value to save

• step – global step for when the value was taken

sparseml.keras.utils.model module

Utils for Keras model

sparseml.keras.utils.model.sparsity(model: tensorflow.python.keras.engine.training.Model)

Retrieve sparsity of a Keras model

Parameters

model – a Keras model

Returns

1. model sparsity, (2) dictionary of layer sparsity

Module contents

Generic code used as utilities and helpers for Keras