org.deeplearning4j.nn.modelimport.keras.utils

Class KerasModelBuilder

java.lang.Object

org.deeplearning4j.nn.modelimport.keras.utils.KerasModelBuilder

All Implemented Interfaces:

Closeable, AutoCloseable, Cloneable

public class KerasModelBuilder
extends Object
implements Cloneable, Closeable

Field Summary

Fields

Modifier and Type	Field and Description
protected KerasModelConfiguration	config
protected KerasLayer.DimOrder	dimOrder
protected boolean	enforceTrainingConfig
protected int[]	inputShape
protected String	modelJson
protected String	modelYaml
protected Hdf5Archive	trainingArchive
protected String	trainingJson
protected String	trainingYaml
protected Hdf5Archive	weightsArchive
protected String	weightsRoot

Constructor Summary

Constructors

Constructor and Description
KerasModelBuilder(KerasModelConfiguration config) KerasModelBuilder constructed from a model configuration.

Method Summary

Modifier and Type	Method and Description
KerasModel	buildModel() Build a KerasModel (corresponding to ComputationGraph) from this model builder.
KerasSequentialModel	buildSequential() Build a KerasSequentialModel (corresponding to MultiLayerNetwork) from this model builder.
void	close() Close all HDF5 archives for this model builder.
KerasModelBuilder	dimOrder(KerasLayer.DimOrder dimOrder) Manually set dim order for Keras model, i.e.
KerasModelBuilder	enforceTrainingConfig(boolean enforceTrainingConfig) Determine whether to enforce loading a training configuration or not.
KerasModelBuilder	inputShape(int[] inputShape) Provide input shape for Keras models that have been compiled without one.
KerasModelBuilder	modelHdf5Filename(String modelHdf5Filename) Set full model HDF5 (architecture, weights and training configuration) by providing the HDF5 filename.
KerasModelBuilder	modelJson(String modelJson) Set model architecture from model JSON string.
KerasModelBuilder	modelJsonFilename(String modelJsonFilename) Set model architecture from file name pointing to model JSON string.
KerasModelBuilder	modelJsonInputStream(InputStream modelJsonInputStream) Set model architecture from input stream of model JSON.
KerasModelBuilder	modelYaml(String modelYaml) Set model architecture from model YAML string.
KerasModelBuilder	modelYamlFilename(String modelYamlFilename) Set model architecture from file name pointing to model YAML string.
KerasModelBuilder	modelYamlInputStream(InputStream modelYamlInputStream) Set model architecture from input stream of model YAML.
KerasModelBuilder	trainingJson(String trainingJson) Provide training configuration as JSON string
KerasModelBuilder	trainingJsonInputStream(InputStream trainingJsonInputStream) Provide training configuration as file input stream from JSON
KerasModelBuilder	trainingYaml(String trainingYaml) Provide training configuration as YAML string
KerasModelBuilder	trainingYamlInputStream(InputStream trainingYamlInputStream) Provide training configuration as file input stream from YAML
KerasModelBuilder	weightsHdf5Filename(String weightsHdf5Filename) Set weights of the model by providing the file name of the corresponding weights HDF5 file.
KerasModelBuilder	weightsHdf5FilenameNoRoot(String weightsHdf5Filename) Set weights of the model by providing the file name of the corresponding weights HDF5 file.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Detail

modelJson

protected String modelJson

modelYaml

protected String modelYaml

trainingJson

protected String trainingJson

trainingYaml

protected String trainingYaml

weightsArchive

protected Hdf5Archive weightsArchive

weightsRoot

protected String weightsRoot

trainingArchive

protected Hdf5Archive trainingArchive

enforceTrainingConfig

protected boolean enforceTrainingConfig

config

protected KerasModelConfiguration config

inputShape

protected int[] inputShape

dimOrder

protected KerasLayer.DimOrder dimOrder

Constructor Detail

KerasModelBuilder

public KerasModelBuilder(KerasModelConfiguration config)

KerasModelBuilder constructed from a model configuration.

Parameters:

config - KerasModelConfiguration

Method Detail

modelJson

public KerasModelBuilder modelJson(String modelJson)

Set model architecture from model JSON string.

Parameters:

modelJson - model as JSON string.

Returns:

Model Builder

modelYaml

public KerasModelBuilder modelYaml(String modelYaml)

Set model architecture from model YAML string.

Parameters:

modelYaml - model as YAML string.

Returns:

Model Builder

modelJsonFilename

public KerasModelBuilder modelJsonFilename(String modelJsonFilename)
                                    throws IOException

Set model architecture from file name pointing to model JSON string.

Parameters:

modelJsonFilename - Name of file containing model JSON string

Returns:

Model Builder

Throws:

IOException - I/O Exception

modelYamlFilename

public KerasModelBuilder modelYamlFilename(String modelYamlFilename)
                                    throws IOException

Set model architecture from file name pointing to model YAML string.

Parameters:

modelYamlFilename - Name of file containing model YAML string

Returns:

Model Builder

Throws:

IOException - I/O Exception

modelJsonInputStream

public KerasModelBuilder modelJsonInputStream(InputStream modelJsonInputStream)
                                       throws IOException

Set model architecture from input stream of model JSON.

Parameters:

modelJsonInputStream - Input stream of model JSON

Returns:

Model builder

Throws:

IOException - I/O exception

modelYamlInputStream

public KerasModelBuilder modelYamlInputStream(InputStream modelYamlInputStream)
                                       throws IOException

Set model architecture from input stream of model YAML.

Parameters:

modelYamlInputStream - Input stream of model YAML

Returns:

Model builder

Throws:

IOException - I/O exception

inputShape

public KerasModelBuilder inputShape(int[] inputShape)

Provide input shape for Keras models that have been compiled without one. DL4J needs this shape information on import to infer shapes of later layers and do shape validation.

Parameters:

inputShape - Input shape as int array

Returns:

Model Builder

trainingJson

public KerasModelBuilder trainingJson(String trainingJson)

Provide training configuration as JSON string

Parameters:

trainingJson - Training JSON string

Returns:

Model builder

trainingYaml

public KerasModelBuilder trainingYaml(String trainingYaml)

Provide training configuration as YAML string

Parameters:

trainingYaml - Training YAML string

Returns:

Model builder

dimOrder

public KerasModelBuilder dimOrder(KerasLayer.DimOrder dimOrder)

Manually set dim order for Keras model, i.e. either TENSORFLOW (channels last) or THEANO (channels first). Dim ordering will be automatically inferred from your model file, so don't tamper with this option unless you're sure what you're doing. Explicitly setting dim ordering can be useful for very old Keras models (before version 1.2), for which inference can be difficult.

Parameters:

dimOrder - Ordering of dimensions (channels first vs. last)

Returns:

Model builder

trainingJsonInputStream

public KerasModelBuilder trainingJsonInputStream(InputStream trainingJsonInputStream)
                                          throws IOException

Provide training configuration as file input stream from JSON

Parameters:

trainingJsonInputStream - Input stream of training JSON string

Returns:

Model builder

Throws:

IOException

trainingYamlInputStream

public KerasModelBuilder trainingYamlInputStream(InputStream trainingYamlInputStream)
                                          throws IOException

Provide training configuration as file input stream from YAML

Parameters:

trainingYamlInputStream - Input stream of training YAML string

Returns:

Model builder

Throws:

IOException

modelHdf5Filename

public KerasModelBuilder modelHdf5Filename(String modelHdf5Filename)
                                    throws UnsupportedKerasConfigurationException,
                                           InvalidKerasConfigurationException,
                                           IOException

Set full model HDF5 (architecture, weights and training configuration) by providing the HDF5 filename.

Parameters:

modelHdf5Filename - File name of HDF5 file containing full model

Returns:

Model builder

Throws:

UnsupportedKerasConfigurationException - Unsupported configuration

InvalidKerasConfigurationException - Invalid configuration

IOException - I/O exception

weightsHdf5FilenameNoRoot

public KerasModelBuilder weightsHdf5FilenameNoRoot(String weightsHdf5Filename)
                                            throws IOException

Set weights of the model by providing the file name of the corresponding weights HDF5 file. The root of the HDF5 group containing weights won't be set by this method.

Parameters:

weightsHdf5Filename - File name of weights HDF5

Returns:

Model builder

Throws:

IOException

weightsHdf5Filename

public KerasModelBuilder weightsHdf5Filename(String weightsHdf5Filename)
                                      throws IOException

Set weights of the model by providing the file name of the corresponding weights HDF5 file. The root of the HDF5 group containing weights will be read and set from the configuration of this model builder instance.

Parameters:

weightsHdf5Filename - File name of weights HDF5

Returns:

Model builder

Throws:

IOException

enforceTrainingConfig

public KerasModelBuilder enforceTrainingConfig(boolean enforceTrainingConfig)

Determine whether to enforce loading a training configuration or not.

Parameters:

enforceTrainingConfig - boolean, read training config or not

Returns:

Model builder

buildModel

public KerasModel buildModel()
                      throws IOException,
                             InvalidKerasConfigurationException,
                             UnsupportedKerasConfigurationException

Build a KerasModel (corresponding to ComputationGraph) from this model builder.

Returns:

KerasModel

Throws:

IOException - I/O exception

InvalidKerasConfigurationException - Invalid configuration

UnsupportedKerasConfigurationException - Unsupported configuration

buildSequential

public KerasSequentialModel buildSequential()
                                     throws IOException,
                                            InvalidKerasConfigurationException,
                                            UnsupportedKerasConfigurationException

Build a KerasSequentialModel (corresponding to MultiLayerNetwork) from this model builder.

Returns:

KerasSequentialModel

Throws:

IOException - I/O exception

InvalidKerasConfigurationException - Invalid configuration

UnsupportedKerasConfigurationException - Unsupported configuration

close

public void close()

Close all HDF5 archives for this model builder.

Specified by:

close in interface Closeable

Specified by:

close in interface AutoCloseable