Package org.deeplearning4j.models.word2vec

Class Word2Vec.Builder

java.lang.Object

org.deeplearning4j.models.sequencevectors.SequenceVectors.Builder<VocabWord>

org.deeplearning4j.models.word2vec.Word2Vec.Builder

Direct Known Subclasses:

ParagraphVectors.Builder

Enclosing class:

Word2Vec

public static class Word2Vec.Builder
extends SequenceVectors.Builder<VocabWord>

Field Summary

Fields

Modifier and Type	Field	Description
protected boolean	allowParallelTokenization
protected LabelAwareIterator	labelAwareIterator
protected SentenceIterator	sentenceIterator
protected TokenizerFactory	tokenizerFactory

Fields inherited from class org.deeplearning4j.models.sequencevectors.SequenceVectors.Builder

batchSize, configuration, elementsLearningAlgorithm, enableScavenger, existingVectors, hugeModelExpected, intersectVectors, iterations, iterator, layerSize, learningRate, learningRateDecayWords, lockFactor, lookupTable, minLearningRate, minWordFrequency, modelUtils, negative, numEpochs, preciseMode, preciseWeightInit, resetModel, sampling, seed, sequenceLearningAlgorithm, STOP, stopWords, trainElementsVectors, trainSequenceVectors, UNK, unknownElement, useAdaGrad, useHierarchicSoftmax, useUnknown, variableWindows, vectorsListeners, vocabCache, vocabLimit, window, workers

Constructor Summary

Constructors

Constructor	Description
Builder()
Builder(@NonNull VectorsConfiguration configuration)

Method Summary

Modifier and Type	Method	Description
Word2Vec.Builder	allowParallelTokenization(boolean allow)	This method enables/disables parallel tokenization.
Word2Vec.Builder	batchSize(int batchSize)	This method defines mini-batch size
Word2Vec	build()	Build SequenceVectors instance with defined settings/options
Word2Vec.Builder	elementsLearningAlgorithm(@NonNull String algorithm)	* Sets specific LearningAlgorithm as Elements Learning Algorithm
Word2Vec.Builder	elementsLearningAlgorithm(@NonNull ElementsLearningAlgorithm<VocabWord> algorithm)	* Sets specific LearningAlgorithm as Elements Learning Algorithm
Word2Vec.Builder	enableScavenger(boolean reallyEnable)	This method ebables/disables periodical vocab truncation during construction Default value: disabled
Word2Vec.Builder	epochs(int numEpochs)	This method defines number of epochs (iterations over whole training corpus) for training
Word2Vec.Builder	intersectModel(@NonNull SequenceVectors vectors, boolean isLocked)
Word2Vec.Builder	iterate(@NonNull SequenceIterator<VocabWord> iterator)	This method used to feed SequenceIterator, that contains training corpus, into ParagraphVectors
Word2Vec.Builder	iterate(@NonNull DocumentIterator iterator)
Word2Vec.Builder	iterate(@NonNull LabelAwareIterator iterator)	This method used to feed LabelAwareIterator, that is usually used
Word2Vec.Builder	iterate(@NonNull SentenceIterator iterator)	This method used to feed SentenceIterator, that contains training corpus, into ParagraphVectors
Word2Vec.Builder	iterations(int iterations)	This method defines number of iterations done for each mini-batch during training
Word2Vec.Builder	layerSize(int layerSize)	This method defines number of dimensions for output vectors
Word2Vec.Builder	learningRate(double learningRate)	This method defines initial learning rate for model training
Word2Vec.Builder	limitVocabularySize(int limit)	This method sets vocabulary limit during construction.
Word2Vec.Builder	lookupTable(@NonNull WeightLookupTable<VocabWord> lookupTable)	This method allows to define external WeightLookupTable to be used
Word2Vec.Builder	minLearningRate(double minLearningRate)	This method defines minimal learning rate value for training
Word2Vec.Builder	minWordFrequency(int minWordFrequency)	This method defines minimal word frequency in training corpus.
Word2Vec.Builder	modelUtils(@NonNull ModelUtils<VocabWord> modelUtils)	Sets ModelUtils that gonna be used as provider for utility methods: similarity(), wordsNearest(), accuracy(), etc
Word2Vec.Builder	negativeSample(double negative)	This method defines whether negative sampling should be used or not PLEASE NOTE: If you're going to use negative sampling, you might want to disable HierarchicSoftmax, which is enabled by default Default value: 0
Word2Vec.Builder	resetModel(boolean reallyReset)	This method defines whether model should be totally wiped out prior building, or not
Word2Vec.Builder	sampling(double sampling)	This method defines whether subsampling should be used or not
Word2Vec.Builder	seed(long randomSeed)	This method defines random seed for random numbers generator
Word2Vec.Builder	setVectorsListeners(@NonNull Collection<VectorsListener<VocabWord>> vectorsListeners)	This method sets VectorsListeners for this SequenceVectors model
Word2Vec.Builder	stopWords(@NonNull Collection<VocabWord> stopList)	This method defines stop words that should be ignored during training
Word2Vec.Builder	stopWords(@NonNull List<String> stopList)	This method defines stop words that should be ignored during training
Word2Vec.Builder	tokenizerFactory(@NonNull TokenizerFactory tokenizerFactory)	This method defines TokenizerFactory to be used for strings tokenization during training PLEASE NOTE: If external VocabCache is used, the same TokenizerFactory should be used to keep derived tokens equal.
Word2Vec.Builder	trainElementsRepresentation(boolean trainElements)	This method is hardcoded to TRUE, since that's whole point of Word2Vec
Word2Vec.Builder	trainSequencesRepresentation(boolean trainSequences)	This method is hardcoded to FALSE, since that's whole point of Word2Vec
Word2Vec.Builder	unknownElement(VocabWord element)	This method allows you to specify SequenceElement that will be used as UNK element, if UNK is used
Word2Vec.Builder	useAdaGrad(boolean reallyUse)	This method defines whether adaptive gradients should be used or not
protected Word2Vec.Builder	useExistingWordVectors(@NonNull WordVectors vec)	This method has no effect for Word2Vec
Word2Vec.Builder	useHierarchicSoftmax(boolean reallyUse)	This method enables/disables Hierarchic softmax Default value: enabled
Word2Vec.Builder	usePreciseMode(boolean reallyUse)
Word2Vec.Builder	usePreciseWeightInit(boolean reallyUse)	If set to true, initial weights for elements/sequences will be derived from elements themself.
Word2Vec.Builder	useUnknown(boolean reallyUse)	This method allows you to specify, if UNK word should be used internally
Word2Vec.Builder	useVariableWindow(int... windows)	This method allows to use variable window size.
Word2Vec.Builder	vocabCache(@NonNull VocabCache<VocabWord> vocabCache)	This method allows to define external VocabCache to be used
Word2Vec.Builder	windowSize(int windowSize)	This method defines context window size
Word2Vec.Builder	workers(int numWorkers)	This method defines maximum number of concurrent threads available for training

Methods inherited from class org.deeplearning4j.models.sequencevectors.SequenceVectors.Builder

presetTables, sequenceLearningAlgorithm, sequenceLearningAlgorithm

Methods inherited from class java.lang.Object

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

Field Detail

sentenceIterator

protected SentenceIterator sentenceIterator

labelAwareIterator

protected LabelAwareIterator labelAwareIterator

tokenizerFactory

protected TokenizerFactory tokenizerFactory

allowParallelTokenization

protected boolean allowParallelTokenization

Constructor Detail

Builder

public Builder()

Builder

public Builder(@NonNull
               @NonNull VectorsConfiguration configuration)

Method Detail

useExistingWordVectors

protected Word2Vec.Builder useExistingWordVectors(@NonNull
                                                  @NonNull WordVectors vec)

This method has no effect for Word2Vec

Overrides:

useExistingWordVectors in class SequenceVectors.Builder<VocabWord>

Parameters:

vec - existing WordVectors model

Returns:

iterate

public Word2Vec.Builder iterate(@NonNull
                                @NonNull DocumentIterator iterator)

iterate

public Word2Vec.Builder iterate(@NonNull
                                @NonNull SentenceIterator iterator)

This method used to feed SentenceIterator, that contains training corpus, into ParagraphVectors

Parameters:

iterator -

Returns:

tokenizerFactory

public Word2Vec.Builder tokenizerFactory(@NonNull
                                         @NonNull TokenizerFactory tokenizerFactory)

This method defines TokenizerFactory to be used for strings tokenization during training PLEASE NOTE: If external VocabCache is used, the same TokenizerFactory should be used to keep derived tokens equal.

Parameters:

tokenizerFactory -

Returns:

iterate

public Word2Vec.Builder iterate(@NonNull
                                @NonNull SequenceIterator<VocabWord> iterator)

This method used to feed SequenceIterator, that contains training corpus, into ParagraphVectors

Overrides:

iterate in class SequenceVectors.Builder<VocabWord>

Parameters:

iterator -

Returns:

iterate

public Word2Vec.Builder iterate(@NonNull
                                @NonNull LabelAwareIterator iterator)

This method used to feed LabelAwareIterator, that is usually used

Parameters:

iterator -

Returns:

batchSize

public Word2Vec.Builder batchSize(int batchSize)

This method defines mini-batch size

Overrides:

batchSize in class SequenceVectors.Builder<VocabWord>

Parameters:

batchSize -

Returns:

iterations

public Word2Vec.Builder iterations(int iterations)

This method defines number of iterations done for each mini-batch during training

Overrides:

iterations in class SequenceVectors.Builder<VocabWord>

Parameters:

iterations -

Returns:

epochs

public Word2Vec.Builder epochs(int numEpochs)

This method defines number of epochs (iterations over whole training corpus) for training

Overrides:

epochs in class SequenceVectors.Builder<VocabWord>

Parameters:

numEpochs -

Returns:

layerSize

public Word2Vec.Builder layerSize(int layerSize)

This method defines number of dimensions for output vectors

Overrides:

layerSize in class SequenceVectors.Builder<VocabWord>

Parameters:

layerSize -

Returns:

learningRate

public Word2Vec.Builder learningRate(double learningRate)

This method defines initial learning rate for model training

Overrides:

learningRate in class SequenceVectors.Builder<VocabWord>

Parameters:

learningRate -

Returns:

minWordFrequency

public Word2Vec.Builder minWordFrequency(int minWordFrequency)

This method defines minimal word frequency in training corpus. All words below this threshold will be removed prior model training

Overrides:

minWordFrequency in class SequenceVectors.Builder<VocabWord>

Parameters:

minWordFrequency -

Returns:

minLearningRate

public Word2Vec.Builder minLearningRate(double minLearningRate)

This method defines minimal learning rate value for training

Overrides:

minLearningRate in class SequenceVectors.Builder<VocabWord>

Parameters:

minLearningRate -

Returns:

resetModel

public Word2Vec.Builder resetModel(boolean reallyReset)

This method defines whether model should be totally wiped out prior building, or not

Overrides:

resetModel in class SequenceVectors.Builder<VocabWord>

Parameters:

reallyReset -

Returns:

limitVocabularySize

public Word2Vec.Builder limitVocabularySize(int limit)

This method sets vocabulary limit during construction. Default value: 0. Means no limit

Overrides:

limitVocabularySize in class SequenceVectors.Builder<VocabWord>

Parameters:

limit -

Returns:

vocabCache

public Word2Vec.Builder vocabCache(@NonNull
                                   @NonNull VocabCache<VocabWord> vocabCache)

This method allows to define external VocabCache to be used

Overrides:

vocabCache in class SequenceVectors.Builder<VocabWord>

Parameters:

vocabCache -

Returns:

lookupTable

public Word2Vec.Builder lookupTable(@NonNull
                                    @NonNull WeightLookupTable<VocabWord> lookupTable)

This method allows to define external WeightLookupTable to be used

Overrides:

lookupTable in class SequenceVectors.Builder<VocabWord>

Parameters:

lookupTable -

Returns:

sampling

public Word2Vec.Builder sampling(double sampling)

This method defines whether subsampling should be used or not

Overrides:

sampling in class SequenceVectors.Builder<VocabWord>

Parameters:

sampling - set > 0 to subsampling argument, or 0 to disable

Returns:

useAdaGrad

public Word2Vec.Builder useAdaGrad(boolean reallyUse)

This method defines whether adaptive gradients should be used or not

Overrides:

useAdaGrad in class SequenceVectors.Builder<VocabWord>

Parameters:

reallyUse -

Returns:

negativeSample

public Word2Vec.Builder negativeSample(double negative)

This method defines whether negative sampling should be used or not PLEASE NOTE: If you're going to use negative sampling, you might want to disable HierarchicSoftmax, which is enabled by default Default value: 0

Overrides:

negativeSample in class SequenceVectors.Builder<VocabWord>

Parameters:

negative - set > 0 as negative sampling argument, or 0 to disable

Returns:

stopWords

public Word2Vec.Builder stopWords(@NonNull
                                  @NonNull List<String> stopList)

This method defines stop words that should be ignored during training

Overrides:

stopWords in class SequenceVectors.Builder<VocabWord>

Parameters:

stopList -

Returns:

trainElementsRepresentation

public Word2Vec.Builder trainElementsRepresentation(boolean trainElements)

This method is hardcoded to TRUE, since that's whole point of Word2Vec

Overrides:

trainElementsRepresentation in class SequenceVectors.Builder<VocabWord>

Parameters:

trainElements -

Returns:

trainSequencesRepresentation

public Word2Vec.Builder trainSequencesRepresentation(boolean trainSequences)

This method is hardcoded to FALSE, since that's whole point of Word2Vec

Overrides:

trainSequencesRepresentation in class SequenceVectors.Builder<VocabWord>

Parameters:

trainSequences -

Returns:

stopWords

public Word2Vec.Builder stopWords(@NonNull
                                  @NonNull Collection<VocabWord> stopList)

This method defines stop words that should be ignored during training

Overrides:

stopWords in class SequenceVectors.Builder<VocabWord>

Parameters:

stopList -

Returns:

windowSize

public Word2Vec.Builder windowSize(int windowSize)

This method defines context window size

Overrides:

windowSize in class SequenceVectors.Builder<VocabWord>

Parameters:

windowSize -

Returns:

seed

public Word2Vec.Builder seed(long randomSeed)

This method defines random seed for random numbers generator

Overrides:

seed in class SequenceVectors.Builder<VocabWord>

Parameters:

randomSeed -

Returns:

workers

public Word2Vec.Builder workers(int numWorkers)

This method defines maximum number of concurrent threads available for training

Overrides:

workers in class SequenceVectors.Builder<VocabWord>

Parameters:

numWorkers -

Returns:

modelUtils

public Word2Vec.Builder modelUtils(@NonNull
                                   @NonNull ModelUtils<VocabWord> modelUtils)

Sets ModelUtils that gonna be used as provider for utility methods: similarity(), wordsNearest(), accuracy(), etc

Overrides:

modelUtils in class SequenceVectors.Builder<VocabWord>

Parameters:

modelUtils - model utils to be used

Returns:

useVariableWindow

public Word2Vec.Builder useVariableWindow(int... windows)

This method allows to use variable window size. In this case, every batch gets processed using one of predefined window sizes

Overrides:

useVariableWindow in class SequenceVectors.Builder<VocabWord>

Parameters:

windows -

Returns:

unknownElement

public Word2Vec.Builder unknownElement(VocabWord element)

This method allows you to specify SequenceElement that will be used as UNK element, if UNK is used

Overrides:

unknownElement in class SequenceVectors.Builder<VocabWord>

Parameters:

element -

Returns:

useUnknown

public Word2Vec.Builder useUnknown(boolean reallyUse)

This method allows you to specify, if UNK word should be used internally

Overrides:

useUnknown in class SequenceVectors.Builder<VocabWord>

Parameters:

reallyUse -

Returns:

setVectorsListeners

public Word2Vec.Builder setVectorsListeners(@NonNull
                                            @NonNull Collection<VectorsListener<VocabWord>> vectorsListeners)

This method sets VectorsListeners for this SequenceVectors model

Overrides:

setVectorsListeners in class SequenceVectors.Builder<VocabWord>

Parameters:

vectorsListeners -

Returns:

elementsLearningAlgorithm

public Word2Vec.Builder elementsLearningAlgorithm(@NonNull
                                                  @NonNull String algorithm)

Description copied from class: SequenceVectors.Builder

* Sets specific LearningAlgorithm as Elements Learning Algorithm

Overrides:

elementsLearningAlgorithm in class SequenceVectors.Builder<VocabWord>

Parameters:

algorithm - fully qualified class name

Returns:

elementsLearningAlgorithm

public Word2Vec.Builder elementsLearningAlgorithm(@NonNull
                                                  @NonNull ElementsLearningAlgorithm<VocabWord> algorithm)

Description copied from class: SequenceVectors.Builder

* Sets specific LearningAlgorithm as Elements Learning Algorithm

Overrides:

elementsLearningAlgorithm in class SequenceVectors.Builder<VocabWord>

Parameters:

algorithm - ElementsLearningAlgorithm implementation

Returns:

allowParallelTokenization

public Word2Vec.Builder allowParallelTokenization(boolean allow)

This method enables/disables parallel tokenization. Default value: TRUE

Parameters:

allow -

Returns:

enableScavenger

public Word2Vec.Builder enableScavenger(boolean reallyEnable)

This method ebables/disables periodical vocab truncation during construction Default value: disabled

Overrides:

enableScavenger in class SequenceVectors.Builder<VocabWord>

Parameters:

reallyEnable -

Returns:

useHierarchicSoftmax

public Word2Vec.Builder useHierarchicSoftmax(boolean reallyUse)

This method enables/disables Hierarchic softmax Default value: enabled

Overrides:

useHierarchicSoftmax in class SequenceVectors.Builder<VocabWord>

Parameters:

reallyUse -

Returns:

usePreciseWeightInit

public Word2Vec.Builder usePreciseWeightInit(boolean reallyUse)

Description copied from class: SequenceVectors.Builder

If set to true, initial weights for elements/sequences will be derived from elements themself. However, this implies additional cycle through input iterator. Default value: FALSE

Overrides:

usePreciseWeightInit in class SequenceVectors.Builder<VocabWord>

Returns:

usePreciseMode

public Word2Vec.Builder usePreciseMode(boolean reallyUse)

Overrides:

usePreciseMode in class SequenceVectors.Builder<VocabWord>

intersectModel

public Word2Vec.Builder intersectModel(@NonNull
                                       @NonNull SequenceVectors vectors,
                                       boolean isLocked)

Overrides:

intersectModel in class SequenceVectors.Builder<VocabWord>

build

public Word2Vec build()

Description copied from class: SequenceVectors.Builder

Build SequenceVectors instance with defined settings/options

Overrides:

build in class SequenceVectors.Builder<VocabWord>

Returns: