org.deeplearning4j.models.word2vec.wordstore

Class VocabularyHolder

java.lang.Object

org.deeplearning4j.models.word2vec.wordstore.VocabularyHolder

All Implemented Interfaces:

java.io.Serializable

public class VocabularyHolder
extends java.lang.Object
implements java.io.Serializable

This class is used as simplifed VocabCache for vocabulary building routines. As soon as vocab is built, all words will be transferred into VocabCache.

See Also:

Serialized Form

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static class	VocabularyHolder.Builder

Constructor Summary

Constructors

Modifier	Constructor and Description
protected	VocabularyHolder() Default constructor
protected	VocabularyHolder(VocabCache<? extends SequenceElement> cache, boolean markAsSpecial) Builds VocabularyHolder from VocabCache.

Method Summary

Modifier and Type	Method and Description
protected void	activateScavenger() This method removes low-frequency words based on their frequency change between activations.
void	addWord(java.lang.String word) Adds new word to vocabulary
void	addWord(VocabularyWord word)
static java.util.List<java.lang.Byte>	arrayToList(byte[] array, int codeLen) This method is used only for VocabCache compatibility purposes
static java.util.List<java.lang.Integer>	arrayToList(int[] array, int codeLen) This method is used only for VocabCache compatibility purposes
static HuffmanNode	buildNode(java.util.List<java.lang.Byte> codes, java.util.List<java.lang.Integer> points, int codeLen, int index)
void	consumeVocabulary(VocabularyHolder holder)
boolean	containsWord(java.lang.String word) Checks vocabulary for the word existance
java.util.Collection<VocabularyWord>	getVocabulary()
VocabularyWord	getVocabularyWordByIdx(java.lang.Integer id)
VocabularyWord	getVocabularyWordByString(java.lang.String word)
void	incrementWordCounter(java.lang.String word) Increments by one number of occurencies of the word in corpus
int	indexOf(java.lang.String word) This method returns index of word in sorted list.
static byte[]	listToArray(java.util.List<java.lang.Byte> code)
static int[]	listToArray(java.util.List<java.lang.Integer> points, int codeLen)
int	numWords()
void	resetWordCounters() This methods reset counters for all words in vocabulary
protected void	setScavengerActivationThreshold(int threshold) This method is needed ONLY for unit tests and should NOT be available in public scope.
long	totalWordsBeyondLimit()
void	transferBackToVocabCache()
void	transferBackToVocabCache(VocabCache cache)
void	transferBackToVocabCache(VocabCache cache, boolean emptyHolder) This method is required for compatibility purposes.
void	truncateVocabulary() The same as truncateVocabulary(this.minWordFrequency)
void	truncateVocabulary(int threshold) All words with frequency below threshold wii be removed
java.util.List<VocabularyWord>	updateHuffmanCodes() build binary tree ordered by counter.
java.util.List<VocabularyWord>	words() Returns sorted list of words in vocabulary.

Methods inherited from class java.lang.Object

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

Constructor Detail

VocabularyHolder

protected VocabularyHolder()

Default constructor

VocabularyHolder

protected VocabularyHolder(@NonNull
                           VocabCache<? extends SequenceElement> cache,
                           boolean markAsSpecial)

Builds VocabularyHolder from VocabCache. Basically we just ignore tokens, and transfer VocabularyWords, supposing that it's already truncated by minWordFrequency. Huffman tree data is ignored and recalculated, due to suspectable flaw in dl4j huffman impl, and it's exsessive memory usage. This code is required for compatibility between dl4j w2v implementation, and standalone w2v

Parameters:

cache -

Method Detail

buildNode

public static HuffmanNode buildNode(java.util.List<java.lang.Byte> codes,
                                    java.util.List<java.lang.Integer> points,
                                    int codeLen,
                                    int index)

transferBackToVocabCache

public void transferBackToVocabCache()

transferBackToVocabCache

public void transferBackToVocabCache(VocabCache cache)

transferBackToVocabCache

public void transferBackToVocabCache(VocabCache cache,
                                     boolean emptyHolder)

This method is required for compatibility purposes. It just transfers vocabulary from VocabHolder into VocabCache

Parameters:

cache -

setScavengerActivationThreshold

protected void setScavengerActivationThreshold(int threshold)

This method is needed ONLY for unit tests and should NOT be available in public scope. It sets the vocab size ratio, at wich dynamic scavenger will be activated

Parameters:

threshold -

arrayToList

public static java.util.List<java.lang.Byte> arrayToList(byte[] array,
                                                         int codeLen)

This method is used only for VocabCache compatibility purposes

Parameters:

array -

codeLen -

Returns:

listToArray

public static byte[] listToArray(java.util.List<java.lang.Byte> code)

listToArray

public static int[] listToArray(java.util.List<java.lang.Integer> points,
                                int codeLen)

arrayToList

public static java.util.List<java.lang.Integer> arrayToList(int[] array,
                                                            int codeLen)

This method is used only for VocabCache compatibility purposes

Parameters:

array -

codeLen -

Returns:

getVocabulary

public java.util.Collection<VocabularyWord> getVocabulary()

getVocabularyWordByString

public VocabularyWord getVocabularyWordByString(java.lang.String word)

getVocabularyWordByIdx

public VocabularyWord getVocabularyWordByIdx(java.lang.Integer id)

containsWord

public boolean containsWord(java.lang.String word)

Checks vocabulary for the word existance

Parameters:

word - to be looked for

Returns:

TRUE of contains, FALSE otherwise

incrementWordCounter

public void incrementWordCounter(java.lang.String word)

Increments by one number of occurencies of the word in corpus

Parameters:

word - whose counter is to be incremented

addWord

public void addWord(java.lang.String word)

Adds new word to vocabulary

Parameters:

word - to be added

addWord

public void addWord(VocabularyWord word)

consumeVocabulary

public void consumeVocabulary(VocabularyHolder holder)

activateScavenger

protected void activateScavenger()

This method removes low-frequency words based on their frequency change between activations. I.e. if word has appeared only once, and it's retained the same frequency over consequence activations, we can assume it can be removed freely

resetWordCounters

public void resetWordCounters()

This methods reset counters for all words in vocabulary

numWords

public int numWords()

Returns:

number of words in vocabulary

truncateVocabulary

public void truncateVocabulary()

The same as truncateVocabulary(this.minWordFrequency)

truncateVocabulary

public void truncateVocabulary(int threshold)

All words with frequency below threshold wii be removed

Parameters:

threshold - exclusive threshold for removal

updateHuffmanCodes

public java.util.List<VocabularyWord> updateHuffmanCodes()

build binary tree ordered by counter. Based on original w2v by google

indexOf

public int indexOf(java.lang.String word)

This method returns index of word in sorted list.

Parameters:

word -

Returns:

words

public java.util.List<VocabularyWord> words()

Returns sorted list of words in vocabulary. Sort is DESCENDING.

Returns:

list of VocabularyWord

totalWordsBeyondLimit

public long totalWordsBeyondLimit()