def load(self,
compile_model: bool = False,
custom_objects: Optional[dict] = None) -> Model:
"""
Loads a Tensorflow model from a TileDB array.
:param compile_model: Boolean. Whether to compile the model after loading or not.
:param custom_objects: Optional dictionary mapping names (strings) to
custom classes or functions to be considered during deserialization.
:return: Model. Tensorflow model.
"""
model_array = tiledb.open(self.uri)
model_array_results = model_array[:]
model_weights = pickle.loads(
model_array_results["model_weights"].item(0))
model_config = json.loads(model_array.meta["model_config"])
architecture = model_config["config"]
model_class = model_config["class_name"]
if model_class == "Sequential":
model = tf.keras.Sequential.from_config(architecture)
elif model_class == "Functional":
model = tf.keras.Model.from_config(architecture)
else:
raise NotImplementedError(
"No support for Subclassed models at the moment. Your "
"model should be either Sequential or Functional.")
model.set_weights(model_weights)
if compile_model:
optimizer_weights = pickle.loads(
model_array_results["optimizer_weights"].item(0))
training_config = json.loads(model_array.meta["training_config"])
# Compile model.
model.compile(**saving_utils.compile_args_from_training_config(
training_config, custom_objects))
saving_utils.try_build_compiled_arguments(model)
# Set optimizer weights.
if optimizer_weights:
try:
model.optimizer._create_all_weights(
model.trainable_variables)
except (NotImplementedError, AttributeError):
logging.warning(
"Error when creating the weights of optimizer {}, making it "
"impossible to restore the saved optimizer state. As a result, "
"your model is starting with a freshly initialized optimizer."
)
try:
model.optimizer.set_weights(optimizer_weights)
except ValueError:
logging.warning("Error in loading the saved optimizer "
"state. As a result, your model is "
"starting with a freshly initialized "
"optimizer.")
return model
def load(path, compile=True, options=None): # pylint: disable=redefined-builtin
"""Loads Keras objects from a SavedModel.
Any Keras layer or model saved to the SavedModel will be loaded back
as Keras objects. Other objects are loaded as regular trackable objects (same
as `tf.saved_model.load`).
Currently, Keras saving/loading only retains the Keras object's weights,
losses, and call function.
The loaded model can be re-compiled, but the original optimizer, compiled loss
functions, and metrics are not retained. This is temporary, and `model.save`
will soon be able to serialize compiled models.
Args:
path: Path to SavedModel.
compile: If true, compile the model after loading it.
options: Optional `tf.saved_model.LoadOptions` object that specifies
options for loading from SavedModel.
Returns:
Object loaded from SavedModel.
"""
# TODO(kathywu): Add saving/loading of optimizer, compiled losses and metrics.
# TODO(kathywu): Add code to load from objects that contain all endpoints
model = tf_load.load_internal(path,
options=options,
loader_cls=KerasObjectLoader)
# pylint: disable=protected-access
if isinstance(model, training_lib.Model) and compile:
# TODO(kathywu): Use compiled objects from SavedModel, instead of
# creating new objects from the training config.
training_config = model._serialized_attributes['metadata'].get(
'training_config', None)
if training_config is not None:
model.compile(**saving_utils.compile_args_from_training_config(
training_config))
saving_utils.try_build_compiled_arguments(model)
else:
logging.warning(
'No training configuration found in save file, so the '
'model was *not* compiled. Compile it manually.')
# pylint: enable=protected-access
# Force variables and resources to initialize.
if not context.executing_eagerly():
sess = backend.get_session() # Variables are initialized by this call.
sess.run(ops.get_collection(ops.GraphKeys.TABLE_INITIALIZERS))
return model
def load(path, compile=True, options=None): # pylint: disable=redefined-builtin
"""Loads Keras objects from a SavedModel.
Any Keras layer or model saved to the SavedModel will be loaded back
as Keras objects. Other objects are loaded as regular trackable objects (same
as `tf.saved_model.load`).
Currently, Keras saving/loading only retains the Keras object's weights,
losses, and call function.
The loaded model can be re-compiled, but the original optimizer, compiled loss
functions, and metrics are not retained. This is temporary, and `model.save`
will soon be able to serialize compiled models.
Args:
path: Path to SavedModel.
compile: If true, compile the model after loading it.
options: Optional `tf.saved_model.LoadOptions` object that specifies
options for loading from SavedModel.
Returns:
Object loaded from SavedModel.
"""
# TODO(kathywu): Add saving/loading of optimizer, compiled losses and metrics.
# TODO(kathywu): Add code to load from objects that contain all endpoints
# Look for metadata file or parse the SavedModel
metadata = saved_metadata_pb2.SavedMetadata()
meta_graph_def = loader_impl.parse_saved_model(path).meta_graphs[0]
object_graph_def = meta_graph_def.object_graph_def
path_to_metadata_pb = os.path.join(path, constants.SAVED_METADATA_PATH)
if gfile.Exists(path_to_metadata_pb):
try:
with gfile.GFile(path_to_metadata_pb, 'rb') as f:
file_content = f.read()
metadata.ParseFromString(file_content)
except message.DecodeError as e:
raise IOError('Cannot parse keras metadata {}: {}.'
.format(path_to_metadata_pb, str(e)))
else:
logging.warning('SavedModel saved prior to TF 2.4 detected when loading '
'Keras model. Please ensure that you are saving the model '
'with model.save() or tf.keras.models.save_model(), *NOT* '
'tf.saved_model.save(). To confirm, there should be a file '
'named "keras_metadata.pb" in the SavedModel directory.')
_read_legacy_metadata(object_graph_def, metadata)
if not metadata.nodes:
# When there are no Keras objects, return the results from the core loader
return tf_load.load(path, options=options)
# Recreate layers and metrics using the info stored in the metadata.
keras_loader = KerasObjectLoader(metadata, object_graph_def)
keras_loader.load_layers(compile=compile)
# Generate a dictionary of all loaded nodes.
nodes_to_load = {'root': None}
for node_id, loaded_node in keras_loader.loaded_nodes.items():
nodes_to_load[keras_loader.get_path(node_id)] = loaded_node
loaded = tf_load.load_partial(path, nodes_to_load, options=options)
# Finalize the loaded layers and remove the extra tracked dependencies.
keras_loader.finalize_objects()
keras_loader.del_tracking()
model = loaded['root']
# pylint: disable=protected-access
if isinstance(model, training_lib.Model) and compile:
# TODO(kathywu): Use compiled objects from SavedModel, instead of
# creating new objects from the training config.
training_config = model._serialized_attributes['metadata'].get(
'training_config', None)
if training_config is not None:
model.compile(**saving_utils.compile_args_from_training_config(
training_config))
saving_utils.try_build_compiled_arguments(model)
else:
logging.warning('No training configuration found in save file, so the '
'model was *not* compiled. Compile it manually.')
# pylint: enable=protected-access
# Force variables and resources to initialize.
if not context.executing_eagerly():
sess = backend.get_session() # Variables are initialized by this call.
sess.run(ops.get_collection(ops.GraphKeys.TABLE_INITIALIZERS))
return model
def load_model_from_hdf5(filepath, custom_objects=None, compile=True): # pylint: disable=redefined-builtin
"""Loads a model saved via `save_model_to_hdf5`.
Arguments:
filepath: One of the following:
- String, path to the saved model
- `h5py.File` object from which to load the model
custom_objects: Optional dictionary mapping names
(strings) to custom classes or functions to be
considered during deserialization.
compile: Boolean, whether to compile the model
after loading.
Returns:
A Keras model instance. If an optimizer was found
as part of the saved model, the model is already
compiled. Otherwise, the model is uncompiled and
a warning will be displayed. When `compile` is set
to False, the compilation is omitted without any
warning.
Raises:
ImportError: if h5py is not available.
ValueError: In case of an invalid savefile.
"""
if h5py is None:
raise ImportError('`load_model` requires h5py.')
if not custom_objects:
custom_objects = {}
opened_new_file = not isinstance(filepath, h5py.File)
if opened_new_file:
f = h5py.File(filepath, mode='r')
else:
f = filepath
model = None
try:
# instantiate model
model_config = f.attrs.get('model_config')
if model_config is None:
raise ValueError('No model found in config file.')
model_config = json_utils.decode(model_config.decode('utf-8'))
model = model_config_lib.model_from_config(
model_config, custom_objects=custom_objects)
# set weights
load_weights_from_hdf5_group(f['model_weights'], model.layers)
if compile:
# instantiate optimizer
training_config = f.attrs.get('training_config')
if training_config is None:
logging.warning(
'No training configuration found in the save file, so '
'the model was *not* compiled. Compile it manually.')
return model
training_config = json_utils.decode(
training_config.decode('utf-8'))
# Compile model.
model.compile(**saving_utils.compile_args_from_training_config(
training_config, custom_objects))
saving_utils.try_build_compiled_arguments(model)
# Set optimizer weights.
if 'optimizer_weights' in f:
try:
model.optimizer._create_all_weights(
model.trainable_variables)
except (NotImplementedError, AttributeError):
logging.warning(
'Error when creating the weights of optimizer {}, making it '
'impossible to restore the saved optimizer state. As a result, '
'your model is starting with a freshly initialized optimizer.'
)
optimizer_weight_values = load_optimizer_weights_from_hdf5_group(
f)
try:
model.optimizer.set_weights(optimizer_weight_values)
except ValueError:
logging.warning('Error in loading the saved optimizer '
'state. As a result, your model is '
'starting with a freshly initialized '
'optimizer.')
finally:
if opened_new_file:
f.close()
return model
def load(path, compile=True, options=None): # pylint: disable=redefined-builtin
"""Loads Keras objects from a SavedModel.
Any Keras layer or model saved to the SavedModel will be loaded back
as Keras objects. Other objects are loaded as regular trackable objects (same
as `tf.saved_model.load`).
Currently, Keras saving/loading only retains the Keras object's weights,
losses, and call function.
The loaded model can be re-compiled, but the original optimizer, compiled loss
functions, and metrics are not retained. This is temporary, and `model.save`
will soon be able to serialize compiled models.
Args:
path: Path to SavedModel.
compile: If true, compile the model after loading it.
options: Optional `tf.saved_model.LoadOptions` object that specifies
options for loading from SavedModel.
Returns:
Object loaded from SavedModel.
"""
# TODO(kathywu): Add saving/loading of optimizer, compiled losses and metrics.
# TODO(kathywu): Add code to load from objects that contain all endpoints
# The Keras metadata file is not yet saved, so create it from the SavedModel.
metadata = saved_metadata_pb2.SavedMetadata()
meta_graph_def = loader_impl.parse_saved_model(path).meta_graphs[0]
object_graph_def = meta_graph_def.object_graph_def
# TODO(kathywu): When the keras metadata file is saved, load it directly
# instead of calling the _read_legacy_metadata function.
_read_legacy_metadata(object_graph_def, metadata)
if not metadata.nodes:
# When there are no Keras objects, return the results from the core loader
return tf_load.load(path, options=options)
# Recreate layers and metrics using the info stored in the metadata.
keras_loader = KerasObjectLoader(metadata, object_graph_def)
keras_loader.load_layers(compile=compile)
# Generate a dictionary of all loaded nodes.
nodes_to_load = {'root': None}
for node_id, loaded_node in keras_loader.loaded_nodes.items():
nodes_to_load[keras_loader.get_path(node_id)] = loaded_node
loaded = tf_load.load_partial(path, nodes_to_load, options=options)
# Finalize the loaded layers and remove the extra tracked dependencies.
keras_loader.finalize_objects()
keras_loader.del_tracking()
model = loaded['root']
# pylint: disable=protected-access
if isinstance(model, training_lib.Model) and compile:
# TODO(kathywu): Use compiled objects from SavedModel, instead of
# creating new objects from the training config.
training_config = model._serialized_attributes['metadata'].get(
'training_config', None)
if training_config is not None:
model.compile(**saving_utils.compile_args_from_training_config(
training_config))
saving_utils.try_build_compiled_arguments(model)
else:
logging.warning('No training configuration found in save file, so the '
'model was *not* compiled. Compile it manually.')
# pylint: enable=protected-access
# Force variables and resources to initialize.
if not context.executing_eagerly():
sess = backend.get_session() # Variables are initialized by this call.
sess.run(ops.get_collection(ops.GraphKeys.TABLE_INITIALIZERS))
return model
def save_model(model,
filepath,
overwrite=True,
include_optimizer=True,
save_format=None,
signatures=None,
options=None,
save_traces=True):
# pylint: disable=line-too-long
"""Saves a model as a TensorFlow SavedModel or HDF5 file.
See the [Serialization and Saving guide](https://keras.io/guides/serialization_and_saving/)
for details.
Usage:
>>> model = tf.keras.Sequential([
... tf.keras.layers.Dense(5, input_shape=(3,)),
... tf.keras.layers.Softmax()])
>>> model.save('/tmp/model')
>>> loaded_model = tf.keras.models.load_model('/tmp/model')
>>> x = tf.random.uniform((10, 3))
>>> assert np.allclose(model.predict(x), loaded_model.predict(x))
The SavedModel and HDF5 file contains:
- the model's configuration (topology)
- the model's weights
- the model's optimizer's state (if any)
Thus models can be reinstantiated in the exact same state, without any of the
code used for model definition or training.
Note that the model weights may have different scoped names after being
loaded. Scoped names include the model/layer names, such as
`"dense_1/kernel:0"`. It is recommended that you use the layer properties to
access specific variables, e.g. `model.get_layer("dense_1").kernel`.
__SavedModel serialization format__
Keras SavedModel uses `tf.saved_model.save` to save the model and all
trackable objects attached to the model (e.g. layers and variables). The model
config, weights, and optimizer are saved in the SavedModel. Additionally, for
every Keras layer attached to the model, the SavedModel stores:
* the config and metadata -- e.g. name, dtype, trainable status
* traced call and loss functions, which are stored as TensorFlow subgraphs.
The traced functions allow the SavedModel format to save and load custom
layers without the original class definition.
You can choose to not save the traced functions by disabling the `save_traces`
option. This will decrease the time it takes to save the model and the
amount of disk space occupied by the output SavedModel. If you enable this
option, then you _must_ provide all custom class definitions when loading
the model. See the `custom_objects` argument in `tf.keras.models.load_model`.
Args:
model: Keras model instance to be saved.
filepath: One of the following:
- String or `pathlib.Path` object, path where to save the model
- `h5py.File` object where to save the model
overwrite: Whether we should overwrite any existing model at the target
location, or instead ask the user with a manual prompt.
include_optimizer: If True, save optimizer's state together.
save_format: Either 'tf' or 'h5', indicating whether to save the model
to Tensorflow SavedModel or HDF5. Defaults to 'tf' in TF 2.X, and 'h5'
in TF 1.X.
signatures: Signatures to save with the SavedModel. Applicable to the 'tf'
format only. Please see the `signatures` argument in
`tf.saved_model.save` for details.
options: (only applies to SavedModel format) `tf.saved_model.SaveOptions`
object that specifies options for saving to SavedModel.
save_traces: (only applies to SavedModel format) When enabled, the
SavedModel will store the function traces for each layer. This
can be disabled, so that only the configs of each layer are stored.
Defaults to `True`. Disabling this will decrease serialization time and
reduce file size, but it requires that all custom layers/models
implement a `get_config()` method.
Raises:
ImportError: If save format is hdf5, and h5py is not available.
"""
# pylint: enable=line-too-long
from tensorflow.python.keras.engine import sequential # pylint: disable=g-import-not-at-top
default_format = 'tf' if tf2.enabled() else 'h5'
save_format = save_format or default_format
filepath = path_to_string(filepath)
# If the user has not already called fit or built the underlying metrics, we
# should do that before saving to ensure the metric names have all
# appropriate name transformations applied.
saving_utils.try_build_compiled_arguments(model)
if (save_format == 'h5'
or (h5py is not None and isinstance(filepath, h5py.File))
or saving_utils.is_hdf5_filepath(filepath)):
# TODO(b/130258301): add utility method for detecting model type.
if (not model._is_graph_network and # pylint:disable=protected-access
not isinstance(model, sequential.Sequential)):
raise NotImplementedError(
'Saving the model to HDF5 format requires the model to be a '
'Functional model or a Sequential model. It does not work for '
'subclassed models, because such models are defined via the body of '
'a Python method, which isn\'t safely serializable. Consider saving '
'to the Tensorflow SavedModel format (by setting save_format="tf") '
'or using `save_weights`.')
hdf5_format.save_model_to_hdf5(model, filepath, overwrite,
include_optimizer)
else:
with generic_utils.SharedObjectSavingScope():
saved_model_save.save(model, filepath, overwrite,
include_optimizer, signatures, options,
save_traces)
def load(
self,
*,
timestamp: Optional[Timestamp] = None,
compile_model: bool = False,
custom_objects: Optional[Mapping[str, Any]] = None,
input_shape: Optional[Tuple[int, ...]] = None,
) -> tf.keras.Model:
"""
Load a Tensorflow model from a TileDB array.
:param timestamp: Range of timestamps to load fragments of the array which live
in the specified time range.
:param compile_model: Whether to compile the model after loading or not.
:param custom_objects: Mapping of names to custom classes or functions to be
considered during deserialization.
:param input_shape: The shape that the custom model expects as input
:return: Tensorflow model.
"""
# TODO: Change timestamp when issue in core is resolved
with tiledb.open(self.uri, ctx=self.ctx,
timestamp=timestamp) as model_array:
model_array_results = model_array[:]
model_config = json.loads(model_array.meta["model_config"])
model_class = model_config["class_name"]
if model_class != "Sequential" and model_class != "Functional":
with generic_utils.SharedObjectLoadingScope():
with generic_utils.CustomObjectScope(custom_objects or {}):
if hasattr(model_config, "decode"):
model_config = model_config.decode("utf-8")
model = model_config_lib.model_from_config(
model_config, custom_objects=custom_objects)
if not model.built:
model.build(input_shape)
# Load weights for layers
self._load_custom_subclassed_model(model, model_array)
else:
cls = (tf.keras.Sequential
if model_class == "Sequential" else tf.keras.Model)
model = cls.from_config(model_config["config"])
model_weights = pickle.loads(
model_array_results["model_weights"].item(0))
model.set_weights(model_weights)
if compile_model:
optimizer_weights = pickle.loads(
model_array_results["optimizer_weights"].item(0))
training_config = json.loads(
model_array.meta["training_config"])
# Compile model.
model.compile(**saving_utils.compile_args_from_training_config(
training_config, custom_objects))
saving_utils.try_build_compiled_arguments(model)
# Set optimizer weights.
if optimizer_weights:
try:
model.optimizer._create_all_weights(
model.trainable_variables)
except (NotImplementedError, AttributeError):
logging.warning(
"Error when creating the weights of optimizer {}, making it "
"impossible to restore the saved optimizer state. As a result, "
"your model is starting with a freshly initialized optimizer."
)
try:
model.optimizer.set_weights(optimizer_weights)
except ValueError:
logging.warning("Error in loading the saved optimizer "
"state. As a result, your model is "
"starting with a freshly initialized "
"optimizer.")
return model
