org.deeplearning4j.models.rntn

Class RNTN

java.lang.Object

org.deeplearning4j.models.rntn.RNTN

All Implemented Interfaces:

Serializable, Cloneable, Layer, Model

public class RNTN
extends Object
implements Layer

Recursive Neural Tensor Network by Socher et. al This is a modified implementation of the sentiment analysis RNTN from Stanford that is intended to work with more general purpose inputs (scene detection with images, labeling series of sentences, among others) This implementation will also be faster in terms of parallelization as well as integration with native matrices/GPUs

Author:

Adam Gibson

See Also:

Serialized Form

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static class	RNTN.Builder

Nested classes/interfaces inherited from interface org.deeplearning4j.nn.api.Layer

Layer.Type

Field Summary

Fields

Modifier and Type	Field and Description
protected String	activationFunction
protected NeuralNetConfiguration	conf
protected Collection<IterationListener>	iterationListeners
protected int	numParameters
protected String	outputActivation
protected org.nd4j.linalg.learning.AdaGrad	paramAdaGrad
protected double	value

Method Summary

Modifier and Type	Method and Description
void	accumulateScore(double accum)
org.nd4j.linalg.api.ndarray.INDArray	activate()
org.nd4j.linalg.api.ndarray.INDArray	activate(org.nd4j.linalg.api.ndarray.INDArray input)
org.nd4j.linalg.api.ndarray.INDArray	activationMean()
Pair<Gradient,Gradient>	backWard(Gradient errors, Gradient deltas, org.nd4j.linalg.api.ndarray.INDArray activation, String previousActivation)
Gradient	backwardGradient(org.nd4j.linalg.api.ndarray.INDArray activation, Gradient errorSignal)
String	basicCategory(String category)
int	batchSize()
Gradient	calcGradient(Gradient layerError, org.nd4j.linalg.api.ndarray.INDArray indArray)
void	clear()
Layer	clone()
NeuralNetConfiguration	conf()
org.nd4j.linalg.api.ndarray.INDArray	derivativeActivation(org.nd4j.linalg.api.ndarray.INDArray input)
Gradient	error(org.nd4j.linalg.api.ndarray.INDArray input)
Gradient	errorSignal(Gradient error, org.nd4j.linalg.api.ndarray.INDArray input)
void	fit()
void	fit(org.nd4j.linalg.api.ndarray.INDArray data)
void	fit(List<Tree> trainingBatch) Trains the network on this mini batch and waits for the training set to complete
List<scala.concurrent.Future<Object>>	fitAsync(List<Tree> trainingBatch) Trains the network on this mini batch and returns a list of futures for each training job
void	forwardPropagateTree(Tree tree) This is the method to call for assigning labels and node vectors to the Tree.
org.nd4j.linalg.api.ndarray.INDArray	getBinaryClassification(String left, String right)
org.nd4j.linalg.api.ndarray.INDArray	getBinaryINDArray(String left, String right)
org.nd4j.linalg.api.ndarray.INDArray	getBinaryTransform(String left, String right)
org.nd4j.linalg.api.ndarray.INDArray	getClassWForNode(Tree node)
org.nd4j.linalg.api.ndarray.INDArray	getFeatureVector(String word)
org.nd4j.linalg.api.ndarray.INDArray	getINDArrayForNode(Tree node)
Collection<IterationListener>	getIterationListeners()
int	getNumParameters()
ConvexOptimizer	getOptimizer()
org.nd4j.linalg.api.ndarray.INDArray	getParam(String param)
org.nd4j.linalg.api.ndarray.INDArray	getParameters()
org.nd4j.linalg.api.ndarray.INDArray	getUnaryClassification(String category)
double	getValue()
org.nd4j.linalg.api.ndarray.INDArray	getValueGradient(List<Tree> trainingBatch)
String	getVocabWord(String word)
org.nd4j.linalg.api.ndarray.INDArray	getWForNode(Tree node)
Gradient	gradient()
Pair<Gradient,Double>	gradientAndScore()
void	initParams()
org.nd4j.linalg.api.ndarray.INDArray	input()
void	iterate(org.nd4j.linalg.api.ndarray.INDArray input)
void	merge(Layer layer, int batchSize)
int	numParams()
List<org.nd4j.linalg.api.ndarray.INDArray>	output(List<Tree> trees) output the prediction probabilities for each tree
org.nd4j.linalg.api.ndarray.INDArray	params()
Map<String,org.nd4j.linalg.api.ndarray.INDArray>	paramTable()
List<Integer>	predict(List<Tree> trees) output the top level labels for each tree
org.nd4j.linalg.api.ndarray.INDArray	preOutput(org.nd4j.linalg.api.ndarray.INDArray x)
org.nd4j.linalg.api.ndarray.INDArray	randomTransformBlock()
org.nd4j.linalg.api.ndarray.INDArray	randomTransformMatrix()
double	score()
void	setConf(NeuralNetConfiguration conf)
void	setIterationListeners(Collection<IterationListener> listeners)
void	setParam(String key, org.nd4j.linalg.api.ndarray.INDArray val)
void	setParams(org.nd4j.linalg.api.ndarray.INDArray params) Set the parameters for the model
void	setParamTable(Map<String,org.nd4j.linalg.api.ndarray.INDArray> paramTable)
void	setScore()
org.nd4j.linalg.api.ndarray.INDArray	transform(org.nd4j.linalg.api.ndarray.INDArray data)
Layer	transpose()
Layer.Type	type()
void	update(Gradient gradient)
void	validateInput()

Methods inherited from class java.lang.Object

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

Field Detail

conf

protected NeuralNetConfiguration conf

iterationListeners

protected Collection<IterationListener> iterationListeners

value

protected double value

activationFunction

protected String activationFunction

outputActivation

protected String outputActivation

paramAdaGrad

protected org.nd4j.linalg.learning.AdaGrad paramAdaGrad

numParameters

protected int numParameters

Method Detail

getIterationListeners

public Collection<IterationListener> getIterationListeners()

Specified by:

getIterationListeners in interface Layer

setIterationListeners

public void setIterationListeners(Collection<IterationListener> listeners)

Specified by:

setIterationListeners in interface Layer

randomTransformMatrix

public org.nd4j.linalg.api.ndarray.INDArray randomTransformMatrix()

randomTransformBlock

public org.nd4j.linalg.api.ndarray.INDArray randomTransformBlock()

fit

public void fit(List<Tree> trainingBatch)

Trains the network on this mini batch and waits for the training set to complete

Parameters:

trainingBatch - the trees to iterate on

fitAsync

public List<scala.concurrent.Future<Object>> fitAsync(List<Tree> trainingBatch)

Trains the network on this mini batch and returns a list of futures for each training job

Parameters:

trainingBatch - the trees to iterate on

getWForNode

public org.nd4j.linalg.api.ndarray.INDArray getWForNode(Tree node)

getINDArrayForNode

public org.nd4j.linalg.api.ndarray.INDArray getINDArrayForNode(Tree node)

getClassWForNode

public org.nd4j.linalg.api.ndarray.INDArray getClassWForNode(Tree node)

getFeatureVector

public org.nd4j.linalg.api.ndarray.INDArray getFeatureVector(String word)

getVocabWord

public String getVocabWord(String word)

basicCategory

public String basicCategory(String category)

getUnaryClassification

public org.nd4j.linalg.api.ndarray.INDArray getUnaryClassification(String category)

getBinaryClassification

public org.nd4j.linalg.api.ndarray.INDArray getBinaryClassification(String left,
                                                                    String right)

getBinaryTransform

public org.nd4j.linalg.api.ndarray.INDArray getBinaryTransform(String left,
                                                               String right)

getBinaryINDArray

public org.nd4j.linalg.api.ndarray.INDArray getBinaryINDArray(String left,
                                                              String right)

forwardPropagateTree

public void forwardPropagateTree(Tree tree)

This is the method to call for assigning labels and node vectors to the Tree. After calling this, each of the non-leaf nodes will have the node vector and the predictions of their classes assigned to that subtree's node.

output

public List<org.nd4j.linalg.api.ndarray.INDArray> output(List<Tree> trees)

output the prediction probabilities for each tree

Parameters:

trees - the trees to predict

Returns:

the prediction probabilities for each tree

predict

public List<Integer> predict(List<Tree> trees)

output the top level labels for each tree

Parameters:

trees - the trees to predict

Returns:

the prediction labels for each tree

setParams

public void setParams(org.nd4j.linalg.api.ndarray.INDArray params)

Set the parameters for the model

Specified by:

setParams in interface Model

Parameters:

params -

getNumParameters

public int getNumParameters()

getParameters

public org.nd4j.linalg.api.ndarray.INDArray getParameters()

getValueGradient

public org.nd4j.linalg.api.ndarray.INDArray getValueGradient(List<Tree> trainingBatch)

getValue

public double getValue()

fit

public void fit()

Specified by:

fit in interface Model

update

public void update(Gradient gradient)

Specified by:

update in interface Model

score

public double score()

Specified by:

score in interface Model

transform

public org.nd4j.linalg.api.ndarray.INDArray transform(org.nd4j.linalg.api.ndarray.INDArray data)

Specified by:

transform in interface Model

params

public org.nd4j.linalg.api.ndarray.INDArray params()

Specified by:

params in interface Model

numParams

public int numParams()

Specified by:

numParams in interface Model

fit

public void fit(org.nd4j.linalg.api.ndarray.INDArray data)

Specified by:

fit in interface Model

iterate

public void iterate(org.nd4j.linalg.api.ndarray.INDArray input)

Specified by:

iterate in interface Model

gradient

public Gradient gradient()

Specified by:

gradient in interface Model

gradientAndScore

public Pair<Gradient,Double> gradientAndScore()

Specified by:

gradientAndScore in interface Model

batchSize

public int batchSize()

Specified by:

batchSize in interface Model

type

public Layer.Type type()

Specified by:

type in interface Layer

error

public Gradient error(org.nd4j.linalg.api.ndarray.INDArray input)

Specified by:

error in interface Layer

derivativeActivation

public org.nd4j.linalg.api.ndarray.INDArray derivativeActivation(org.nd4j.linalg.api.ndarray.INDArray input)

Specified by:

derivativeActivation in interface Layer

calcGradient

public Gradient calcGradient(Gradient layerError,
                             org.nd4j.linalg.api.ndarray.INDArray indArray)

Specified by:

calcGradient in interface Layer

errorSignal

public Gradient errorSignal(Gradient error,
                            org.nd4j.linalg.api.ndarray.INDArray input)

Specified by:

errorSignal in interface Layer

backwardGradient

public Gradient backwardGradient(org.nd4j.linalg.api.ndarray.INDArray activation,
                                 Gradient errorSignal)

Specified by:

backwardGradient in interface Layer

merge

public void merge(Layer layer,
                  int batchSize)

Specified by:

merge in interface Layer

getParam

public org.nd4j.linalg.api.ndarray.INDArray getParam(String param)

Specified by:

getParam in interface Model

initParams

public void initParams()

Specified by:

initParams in interface Model

paramTable

public Map<String,org.nd4j.linalg.api.ndarray.INDArray> paramTable()

Specified by:

paramTable in interface Model

setParamTable

public void setParamTable(Map<String,org.nd4j.linalg.api.ndarray.INDArray> paramTable)

Specified by:

setParamTable in interface Model

setParam

public void setParam(String key,
                     org.nd4j.linalg.api.ndarray.INDArray val)

Specified by:

setParam in interface Model

clear

public void clear()

Specified by:

clear in interface Model

activationMean

public org.nd4j.linalg.api.ndarray.INDArray activationMean()

Specified by:

activationMean in interface Layer

conf

public NeuralNetConfiguration conf()

Specified by:

conf in interface Model

preOutput

public org.nd4j.linalg.api.ndarray.INDArray preOutput(org.nd4j.linalg.api.ndarray.INDArray x)

Specified by:

preOutput in interface Layer

activate

public org.nd4j.linalg.api.ndarray.INDArray activate()

Specified by:

activate in interface Layer

activate

public org.nd4j.linalg.api.ndarray.INDArray activate(org.nd4j.linalg.api.ndarray.INDArray input)

Specified by:

activate in interface Layer

transpose

public Layer transpose()

Specified by:

transpose in interface Layer

clone

public Layer clone()

Specified by:

clone in interface Layer

Overrides:

clone in class Object

backWard

public Pair<Gradient,Gradient> backWard(Gradient errors,
                                        Gradient deltas,
                                        org.nd4j.linalg.api.ndarray.INDArray activation,
                                        String previousActivation)

Specified by:

backWard in interface Layer

setScore

public void setScore()

Specified by:

setScore in interface Model

accumulateScore

public void accumulateScore(double accum)

Specified by:

accumulateScore in interface Model

setConf

public void setConf(NeuralNetConfiguration conf)

Specified by:

setConf in interface Model

input

public org.nd4j.linalg.api.ndarray.INDArray input()

Specified by:

input in interface Model

validateInput

public void validateInput()

Specified by:

validateInput in interface Model

getOptimizer

public ConvexOptimizer getOptimizer()

Specified by:

getOptimizer in interface Model