org.deeplearning4j.models.paragraphvectors

Class ParagraphVectors

java.lang.Object

org.deeplearning4j.models.embeddings.wordvectors.WordVectorsImpl<T>

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

org.deeplearning4j.models.word2vec.Word2Vec

org.deeplearning4j.models.paragraphvectors.ParagraphVectors

All Implemented Interfaces:

Serializable, WordVectors, org.deeplearning4j.nn.weights.embeddings.EmbeddingInitializer

public class ParagraphVectors
extends Word2Vec

Basic ParagraphVectors (aka Doc2Vec) implementation for DL4j, as wrapper over SequenceVectors

Author:

[email protected]

See Also:

Serialized Form

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
class	ParagraphVectors.BlindInferenceCallable
static class	ParagraphVectors.Builder
class	ParagraphVectors.InferenceCallable

Nested classes/interfaces inherited from class org.deeplearning4j.models.sequencevectors.SequenceVectors

SequenceVectors.AsyncSequencer

Field Summary

Fields

Modifier and Type	Field and Description
protected AtomicLong	countFinished
protected AtomicLong	countSubmitted
protected org.threadly.concurrent.PriorityScheduler	inferenceExecutor
protected Object	inferenceLocker
protected LabelAwareIterator	labelAwareIterator
protected List<VocabWord>	labelsList
protected org.nd4j.linalg.api.ndarray.INDArray	labelsMatrix
protected LabelsSource	labelsSource
protected boolean	normalizedLabels

Fields inherited from class org.deeplearning4j.models.word2vec.Word2Vec

sentenceIter, tokenizerFactory

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

configuration, configured, elementsLearningAlgorithm, enableScavenger, eventListeners, existingModel, intersectModel, iterator, lockFactor, log, scoreElements, scoreSequences, sequenceLearningAlgorithm, unknownElement, vocabLimit

Fields inherited from class org.deeplearning4j.models.embeddings.wordvectors.WordVectorsImpl

batchSize, DEFAULT_UNK, layerSize, learningRate, learningRateDecayWords, lookupTable, minLearningRate, minWordFrequency, modelUtils, negative, numEpochs, numIterations, resetModel, sampling, seed, stopWords, trainElementsVectors, trainSequenceVectors, useAdeGrad, useUnknown, variableWindows, vocab, window, workers

Constructor Summary

Constructors

Modifier	Constructor and Description
protected	ParagraphVectors()

Method Summary

Modifier and Type	Method and Description
void	extractLabels()
void	fit() Starts training over
static ParagraphVectors	fromJson(String jsonString)
org.nd4j.linalg.api.ndarray.INDArray	inferVector(LabelledDocument document) This method calculates inferred vector for given document, with default parameters for learning rate and iterations
org.nd4j.linalg.api.ndarray.INDArray	inferVector(LabelledDocument document, double learningRate, double minLearningRate, int iterations) This method calculates inferred vector for given document
org.nd4j.linalg.api.ndarray.INDArray	inferVector(List<VocabWord> document) This method calculates inferred vector for given list of words, with default parameters for learning rate and iterations
org.nd4j.linalg.api.ndarray.INDArray	inferVector(List<VocabWord> document, double learningRate, double minLearningRate, int iterations) This method calculates inferred vector for given document
org.nd4j.linalg.api.ndarray.INDArray	inferVector(String text) This method calculates inferred vector for given text, with default parameters for learning rate and iterations
org.nd4j.linalg.api.ndarray.INDArray	inferVector(String text, double learningRate, double minLearningRate, int iterations) This method calculates inferred vector for given text
Future<org.nd4j.linalg.primitives.Pair<String,org.nd4j.linalg.api.ndarray.INDArray>>	inferVectorBatched(LabelledDocument document) This method implements batched inference, based on Java Future parallelism model.
List<org.nd4j.linalg.api.ndarray.INDArray>	inferVectorBatched(List<String> documents) This method does inference on a given List<String>
Future<org.nd4j.linalg.api.ndarray.INDArray>	inferVectorBatched(String document) This method implements batched inference, based on Java Future parallelism model.
protected void	initInference()
Collection<String>	nearestLabels(Collection<VocabWord> document, int topN) This method returns top N labels nearest to specified set of vocab words
Collection<String>	nearestLabels(org.nd4j.linalg.api.ndarray.INDArray labelVector, int topN) This method returns top N labels nearest to specified features vector
Collection<String>	nearestLabels(LabelledDocument document, int topN) This method returns top N labels nearest to specified document
Collection<String>	nearestLabels(String rawText, int topN) This method returns top N labels nearest to specified text
String	predict(LabelledDocument document) This method predicts label of the document.
String	predict(List<VocabWord> document) This method predicts label of the document.
String	predict(String rawText) Deprecated.
Collection<String>	predictSeveral(LabelledDocument document, int limit) Predict several labels based on the document.
Collection<String>	predictSeveral(List<VocabWord> document, int limit) Predict several labels based on the document.
Collection<String>	predictSeveral(String rawText, int limit) Predict several labels based on the document.
protected void	reassignExistingModel()
void	setSequenceIterator(SequenceIterator<VocabWord> iterator) This method defines SequenceIterator instance, that will be used as training corpus source.
double	similarityToLabel(LabelledDocument document, String label) This method returns similarity of the document to specific label, based on mean value
double	similarityToLabel(List<VocabWord> document, String label) This method returns similarity of the document to specific label, based on mean value
double	similarityToLabel(String rawText, String label) Deprecated.
String	toJson()

Methods inherited from class org.deeplearning4j.models.word2vec.Word2Vec

setSentenceIterator, setTokenizerFactory

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

buildVocab, getElementsScore, getSequencesScore, getUNK, getWordVectorMatrix, initLearners, setUNK, trainSequence

Methods inherited from class org.deeplearning4j.models.embeddings.wordvectors.WordVectorsImpl

accuracy, getLayerSize, getWordVector, getWordVectorMatrixNormalized, getWordVectors, getWordVectorsMean, hasWord, indexOf, jsonSerializable, loadWeightsInto, lookupTable, outOfVocabularySupported, setLookupTable, setModelUtils, setVocab, similarity, similarWordsInVocabTo, update, update, vectorSize, vocab, vocabSize, wordsNearest, wordsNearest, wordsNearest, wordsNearestSum, wordsNearestSum, wordsNearestSum

Methods inherited from class java.lang.Object

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

Methods inherited from interface org.deeplearning4j.models.embeddings.wordvectors.WordVectors

accuracy, getWordVector, getWordVectorMatrixNormalized, getWordVectors, getWordVectorsMean, hasWord, indexOf, lookupTable, outOfVocabularySupported, setModelUtils, similarity, similarWordsInVocabTo, vocab, wordsNearest, wordsNearest, wordsNearest, wordsNearestSum, wordsNearestSum, wordsNearestSum

Methods inherited from interface org.deeplearning4j.nn.weights.embeddings.EmbeddingInitializer

jsonSerializable, loadWeightsInto, vectorSize, vocabSize

Field Detail

labelsSource

protected LabelsSource labelsSource

labelAwareIterator

protected transient LabelAwareIterator labelAwareIterator

labelsMatrix

protected org.nd4j.linalg.api.ndarray.INDArray labelsMatrix

labelsList

protected List<VocabWord> labelsList

normalizedLabels

protected boolean normalizedLabels

inferenceLocker

protected final transient Object inferenceLocker

inferenceExecutor

protected transient org.threadly.concurrent.PriorityScheduler inferenceExecutor

countSubmitted

protected transient AtomicLong countSubmitted

countFinished

protected transient AtomicLong countFinished

Constructor Detail

ParagraphVectors

protected ParagraphVectors()

Method Detail

initInference

protected void initInference()

predict

@Deprecated
public String predict(String rawText)

Deprecated.

This method takes raw text, applies tokenizer, and returns most probable label

Parameters:

rawText -

Returns:

setSequenceIterator

public void setSequenceIterator(@NonNull
                                SequenceIterator<VocabWord> iterator)

This method defines SequenceIterator instance, that will be used as training corpus source. Main difference with other iterators here: it allows you to pass already tokenized Sequence for training

Overrides:

setSequenceIterator in class Word2Vec

Parameters:

iterator -

predict

public String predict(LabelledDocument document)

This method predicts label of the document. Computes a similarity wrt the mean of the representation of words in the document

Parameters:

document - the document

Returns:

the word distances for each label

extractLabels

public void extractLabels()

inferVector

public org.nd4j.linalg.api.ndarray.INDArray inferVector(String text,
                                                        double learningRate,
                                                        double minLearningRate,
                                                        int iterations)

This method calculates inferred vector for given text

Parameters:

text -

Returns:

reassignExistingModel

protected void reassignExistingModel()

inferVector

public org.nd4j.linalg.api.ndarray.INDArray inferVector(LabelledDocument document,
                                                        double learningRate,
                                                        double minLearningRate,
                                                        int iterations)

This method calculates inferred vector for given document

Parameters:

document -

Returns:

inferVector

public org.nd4j.linalg.api.ndarray.INDArray inferVector(@NonNull
                                                        List<VocabWord> document,
                                                        double learningRate,
                                                        double minLearningRate,
                                                        int iterations)

This method calculates inferred vector for given document

Parameters:

document -

Returns:

inferVector

public org.nd4j.linalg.api.ndarray.INDArray inferVector(String text)

This method calculates inferred vector for given text, with default parameters for learning rate and iterations

Parameters:

text -

Returns:

inferVector

public org.nd4j.linalg.api.ndarray.INDArray inferVector(LabelledDocument document)

This method calculates inferred vector for given document, with default parameters for learning rate and iterations

Parameters:

document -

Returns:

inferVector

public org.nd4j.linalg.api.ndarray.INDArray inferVector(@NonNull
                                                        List<VocabWord> document)

This method calculates inferred vector for given list of words, with default parameters for learning rate and iterations

Parameters:

document -

Returns:

inferVectorBatched

public Future<org.nd4j.linalg.primitives.Pair<String,org.nd4j.linalg.api.ndarray.INDArray>> inferVectorBatched(@NonNull
                                                                                                               LabelledDocument document)

This method implements batched inference, based on Java Future parallelism model. PLEASE NOTE: In order to use this method, LabelledDocument being passed in should have Id field defined.

Parameters:

document -

Returns:

inferVectorBatched

public Future<org.nd4j.linalg.api.ndarray.INDArray> inferVectorBatched(@NonNull
                                                                       String document)

This method implements batched inference, based on Java Future parallelism model. PLEASE NOTE: This method will return you Future<INDArray>, so tracking relation between document and INDArray will be your responsibility

Parameters:

document -

Returns:

inferVectorBatched

public List<org.nd4j.linalg.api.ndarray.INDArray> inferVectorBatched(@NonNull
                                                                     List<String> documents)

This method does inference on a given List<String>

Parameters:

documents -

Returns:

INDArrays in the same order as input texts

predict

public String predict(List<VocabWord> document)

This method predicts label of the document. Computes a similarity wrt the mean of the representation of words in the document

Parameters:

document - the document

Returns:

the word distances for each label

predictSeveral

public Collection<String> predictSeveral(@NonNull
                                         LabelledDocument document,
                                         int limit)

Predict several labels based on the document. Computes a similarity wrt the mean of the representation of words in the document

Parameters:

document - raw text of the document

Returns:

possible labels in descending order

predictSeveral

public Collection<String> predictSeveral(String rawText,
                                         int limit)

Predict several labels based on the document. Computes a similarity wrt the mean of the representation of words in the document

Parameters:

rawText - raw text of the document

Returns:

possible labels in descending order

predictSeveral

public Collection<String> predictSeveral(List<VocabWord> document,
                                         int limit)

Predict several labels based on the document. Computes a similarity wrt the mean of the representation of words in the document

Parameters:

document - the document

Returns:

possible labels in descending order

nearestLabels

public Collection<String> nearestLabels(LabelledDocument document,
                                        int topN)

This method returns top N labels nearest to specified document

Parameters:

document -

topN -

Returns:

nearestLabels

public Collection<String> nearestLabels(@NonNull
                                        String rawText,
                                        int topN)

This method returns top N labels nearest to specified text

Parameters:

rawText -

topN -

Returns:

nearestLabels

public Collection<String> nearestLabels(@NonNull
                                        Collection<VocabWord> document,
                                        int topN)

This method returns top N labels nearest to specified set of vocab words

Parameters:

document -

topN -

Returns:

nearestLabels

public Collection<String> nearestLabels(org.nd4j.linalg.api.ndarray.INDArray labelVector,
                                        int topN)

This method returns top N labels nearest to specified features vector

Parameters:

labelVector -

topN -

Returns:

similarityToLabel

@Deprecated
public double similarityToLabel(String rawText,
                                            String label)

Deprecated.

This method returns similarity of the document to specific label, based on mean value

Parameters:

rawText -

label -

Returns:

fit

public void fit()

Description copied from class: SequenceVectors

Starts training over

Overrides:

fit in class SequenceVectors<VocabWord>

similarityToLabel

public double similarityToLabel(LabelledDocument document,
                                String label)

This method returns similarity of the document to specific label, based on mean value

Parameters:

document -

label -

Returns:

similarityToLabel

public double similarityToLabel(List<VocabWord> document,
                                String label)

This method returns similarity of the document to specific label, based on mean value

Parameters:

document -

label -

Returns:

toJson

public String toJson()
              throws org.nd4j.shade.jackson.core.JsonProcessingException

Overrides:

toJson in class Word2Vec

Throws:

org.nd4j.shade.jackson.core.JsonProcessingException

fromJson

public static ParagraphVectors fromJson(String jsonString)
                                 throws IOException

Throws:

IOException