Package org.deeplearning4j.nn.params

Class PretrainParamInitializer

java.lang.Object

org.deeplearning4j.nn.params.DefaultParamInitializer

org.deeplearning4j.nn.params.PretrainParamInitializer

All Implemented Interfaces:

ParamInitializer

public class PretrainParamInitializer
extends DefaultParamInitializer

Pretrain weight initializer. Has the visible bias as well as hidden and weight matrix.

Author:

Adam Gibson

Field Summary

Fields

Modifier and Type	Field	Description
static String	VISIBLE_BIAS_KEY

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

BIAS_KEY, GAIN_KEY, WEIGHT_KEY

Constructor Summary

Constructors

Constructor	Description
PretrainParamInitializer()

Method Summary

Modifier and Type	Method	Description
protected INDArray	createVisibleBias(NeuralNetConfiguration conf, INDArray visibleBiasView, boolean initializeParameters)
Map<String,INDArray>	getGradientsFromFlattened(NeuralNetConfiguration conf, INDArray gradientView)	Return a map of gradients (in their standard non-flattened representation), taken from the flattened (row vector) gradientView array.
static PretrainParamInitializer	getInstance()
Map<String,INDArray>	init(NeuralNetConfiguration conf, INDArray paramsView, boolean initializeParams)	Initialize the parameters
long	numParams(NeuralNetConfiguration conf)

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

biasKeys, createBias, createBias, createGain, createGain, createWeightMatrix, createWeightMatrix, hasBias, hasLayerNorm, isBiasParam, isWeightParam, numParams, paramKeys, weightKeys

Methods inherited from class java.lang.Object

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

Field Detail

VISIBLE_BIAS_KEY

public static final String VISIBLE_BIAS_KEY

See Also:

Constant Field Values

Constructor Detail

PretrainParamInitializer

public PretrainParamInitializer()

Method Detail

getInstance

public static PretrainParamInitializer getInstance()

numParams

public long numParams(NeuralNetConfiguration conf)

Specified by:

numParams in interface ParamInitializer

Overrides:

numParams in class DefaultParamInitializer

init

public Map<String,INDArray> init(NeuralNetConfiguration conf,
                                       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)

createVisibleBias

protected INDArray createVisibleBias(NeuralNetConfiguration conf,
                                     INDArray visibleBiasView,
                                     boolean initializeParameters)

getGradientsFromFlattened

public Map<String,INDArray> getGradientsFromFlattened(NeuralNetConfiguration conf,
                                                            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 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