simpleml.save_patterns.serializers.keras

Module for Keras save patterns

Module Contents

Classes

KerasH5Serializer

Uses Keras H5 serialization (legacy behavior)

KerasPersistenceMethods

Base class for internal Keras serialization/deserialization options

KerasSavedModelSerializer

Uses Tensorflow SavedModel serialization

Attributes

__author__

simpleml.save_patterns.serializers.keras.__author__ = Elisha Yadgaran[source]
class simpleml.save_patterns.serializers.keras.KerasH5Serializer[source]

Bases: simpleml.save_patterns.base.BaseSerializer

Uses Keras H5 serialization (legacy behavior)

Output is a single file

static deserialize(filepath, source_directory='system_temp', **kwargs)[source]
Parameters

  • filepath (str) –

  • source_directory (str) –

Return type

Dict[str, Any]

static serialize(obj, filepath, format_directory=HDF5_DIRECTORY, format_extension='.h5', destination_directory='system_temp', **kwargs)[source]
Parameters

  • obj (Any) –

  • filepath (str) –

  • format_directory (str) –

  • format_extension (str) –

  • destination_directory (str) –

Return type

Dict[str, str]

class simpleml.save_patterns.serializers.keras.KerasPersistenceMethods[source]

Bases: object

Base class for internal Keras serialization/deserialization options

static load_model(filepath, **kwargs)[source]

Loads a Keras object from the filesystem.

Parameters

filepath (str) –

Return type

Any

static load_weights(model, filepath, **kwargs)[source]

Loads a Keras object from the filesystem.

Parameters

  • model (Any) –

  • filepath (str) –

Return type

Any

static save_model(model, filepath, overwrite=True, **kwargs)[source]

Serializes an object to the filesystem in Keras native format.

Parameters

  • overwrite (bool) – Boolean indicating whether to first check if object is already serialized. Defaults to not checking, but can be leverage by implementations that want the same artifact in multiple places

  • model (Any) –

  • filepath (str) –

Return type

None

static save_weights(model, filepath, overwrite=True, **kwargs)[source]

Serializes an object to the filesystem in Keras native format.

Parameters

  • overwrite (bool) – Boolean indicating whether to first check if object is already serialized. Defaults to not checking, but can be leverage by implementations that want the same artifact in multiple places

  • model (Any) –

  • filepath (str) –

Return type

None

class simpleml.save_patterns.serializers.keras.KerasSavedModelSerializer[source]

Bases: simpleml.save_patterns.base.BaseSerializer

Uses Tensorflow SavedModel serialization

Output is a folder with assets keras_metadata.pb saved_model.pb variables

static deserialize(filepath, source_directory='system_temp', **kwargs)[source]
Parameters

  • filepath (str) –

  • source_directory (str) –

Return type

Dict[str, Any]

static serialize(obj, filepath, format_directory=TENSORFLOW_SAVED_MODEL_DIRECTORY, format_extension='.savedModel', destination_directory='system_temp', **kwargs)[source]
Parameters

  • obj (Any) –

  • filepath (str) –

  • format_directory (str) –

  • format_extension (str) –

  • destination_directory (str) –

Return type

Dict[str, str]