org.deeplearning4j.nn.params

Class VariationalAutoencoderParamInitializer

java.lang.Object

org.deeplearning4j.nn.params.DefaultParamInitializer

org.deeplearning4j.nn.params.VariationalAutoencoderParamInitializer

All Implemented Interfaces:

ParamInitializer

public class VariationalAutoencoderParamInitializer
extends DefaultParamInitializer

Parameter initializer for the Variational Autoencoder model. See: Kingma & Welling, 2013: Auto-Encoding Variational Bayes - https://arxiv.org/abs/1312.6114

Field Summary

Fields

Modifier and Type	Field and Description
static java.lang.String	BIAS_KEY_SUFFIX
static java.lang.String	PXZ_B Key for bias parameters connecting the last decoder layer and p(data|z) (according to whatever ReconstructionDistribution is set for the VAE)
static java.lang.String	PXZ_PREFIX
static java.lang.String	PXZ_W Key for weight parameters connecting the last decoder layer and p(data|z) (according to whatever ReconstructionDistribution is set for the VAE)
static java.lang.String	PZX_LOGSTD2_B Key for bias parameters for log(sigma^2) in p(z|data)
static java.lang.String	PZX_LOGSTD2_PREFIX
static java.lang.String	PZX_LOGSTD2_W Key for weight parameters connecting the last encoder layer and the log(sigma^2) values for p(z|data)
static java.lang.String	PZX_MEAN_B Key for bias parameters for the mean values for p(z|data)
static java.lang.String	PZX_MEAN_PREFIX
static java.lang.String	PZX_MEAN_W Key for weight parameters connecting the last encoder layer and the mean values for p(z|data)
static java.lang.String	PZX_PREFIX
static java.lang.String	WEIGHT_KEY_SUFFIX

Fields inherited from class org.deeplearning4j.nn.params.DefaultParamInitializer

BIAS_KEY, WEIGHT_KEY

Constructor Summary

Constructors

Constructor and Description
VariationalAutoencoderParamInitializer()

Method Summary

Modifier and Type	Method and Description
java.util.Map<java.lang.String,org.nd4j.linalg.api.ndarray.INDArray>	getGradientsFromFlattened(NeuralNetConfiguration conf, org.nd4j.linalg.api.ndarray.INDArray gradientView) Return a map of gradients (in their standard non-flattened representation), taken from the flattened (row vector) gradientView array.
static VariationalAutoencoderParamInitializer	getInstance()
java.util.Map<java.lang.String,org.nd4j.linalg.api.ndarray.INDArray>	init(NeuralNetConfiguration conf, org.nd4j.linalg.api.ndarray.INDArray paramsView, boolean initializeParams) Initialize the parameters
int	numParams(NeuralNetConfiguration conf)

Methods inherited from class org.deeplearning4j.nn.params.DefaultParamInitializer

createBias, createBias, createWeightMatrix, createWeightMatrix

Methods inherited from class java.lang.Object

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

Field Detail

WEIGHT_KEY_SUFFIX

public static final java.lang.String WEIGHT_KEY_SUFFIX

See Also:

Constant Field Values

BIAS_KEY_SUFFIX

public static final java.lang.String BIAS_KEY_SUFFIX

See Also:

Constant Field Values

PZX_PREFIX

public static final java.lang.String PZX_PREFIX

See Also:

Constant Field Values

PZX_MEAN_PREFIX

public static final java.lang.String PZX_MEAN_PREFIX

See Also:

Constant Field Values

PZX_LOGSTD2_PREFIX

public static final java.lang.String PZX_LOGSTD2_PREFIX

See Also:

Constant Field Values

PZX_MEAN_W

public static final java.lang.String PZX_MEAN_W

Key for weight parameters connecting the last encoder layer and the mean values for p(z|data)

See Also:

Constant Field Values

PZX_MEAN_B

public static final java.lang.String PZX_MEAN_B

Key for bias parameters for the mean values for p(z|data)

See Also:

Constant Field Values

PZX_LOGSTD2_W

public static final java.lang.String PZX_LOGSTD2_W

Key for weight parameters connecting the last encoder layer and the log(sigma^2) values for p(z|data)

See Also:

Constant Field Values

PZX_LOGSTD2_B

public static final java.lang.String PZX_LOGSTD2_B

Key for bias parameters for log(sigma^2) in p(z|data)

See Also:

Constant Field Values

PXZ_PREFIX

public static final java.lang.String PXZ_PREFIX

See Also:

Constant Field Values

PXZ_W

public static final java.lang.String PXZ_W

Key for weight parameters connecting the last decoder layer and p(data|z) (according to whatever ReconstructionDistribution is set for the VAE)

See Also:

Constant Field Values

PXZ_B

public static final java.lang.String PXZ_B

Key for bias parameters connecting the last decoder layer and p(data|z) (according to whatever ReconstructionDistribution is set for the VAE)

See Also:

Constant Field Values

Constructor Detail

VariationalAutoencoderParamInitializer

public VariationalAutoencoderParamInitializer()

Method Detail

getInstance

public static VariationalAutoencoderParamInitializer getInstance()

numParams

public int numParams(NeuralNetConfiguration conf)

Specified by:

numParams in interface ParamInitializer

Overrides:

numParams in class DefaultParamInitializer

init

public java.util.Map<java.lang.String,org.nd4j.linalg.api.ndarray.INDArray> init(NeuralNetConfiguration conf,
                                                                                 org.nd4j.linalg.api.ndarray.INDArray paramsView,
                                                                                 boolean initializeParams)

Description copied from interface: ParamInitializer

Initialize the parameters

Specified by:

init in interface ParamInitializer

Overrides:

init in class DefaultParamInitializer

Parameters:

conf - the configuration

paramsView - a view of the full network (backprop) parameters

initializeParams - if true: initialize the parameters according to the configuration. If false: don't modify the values in the paramsView array (but do select out the appropriate subset, reshape etc as required)

Returns:

Map of parameters keyed by type (view of the 'paramsView' array)

getGradientsFromFlattened

public java.util.Map<java.lang.String,org.nd4j.linalg.api.ndarray.INDArray> getGradientsFromFlattened(NeuralNetConfiguration conf,
                                                                                                      org.nd4j.linalg.api.ndarray.INDArray gradientView)

Description copied from interface: ParamInitializer

Return a map of gradients (in their standard non-flattened representation), taken from the flattened (row vector) gradientView array. The idea is that operates in exactly the same way as the the paramsView does in #init(Map, NeuralNetConfiguration, INDArray); thus the position in the view (and, the array orders) must match those of the parameters

Specified by:

getGradientsFromFlattened in interface ParamInitializer

Overrides:

getGradientsFromFlattened in class DefaultParamInitializer

Parameters:

conf - Configuration

gradientView - The flattened gradients array, as a view of the larger array

Returns:

A map containing an array by parameter type, that is a view of the full network gradients array