Example usage for org.deeplearning4j.models.word2vec.wordstore.inmemory AbstractCache addToken

Introduction

In this page you can find the example usage for org.deeplearning4j.models.word2vec.wordstore.inmemory AbstractCache addToken.

Prototype

@Override
public boolean addToken(T element) 

Source Link

Document

This method adds specified SequenceElement to vocabulary

Usage

From source file:edu.umd.umiacs.clip.tools.scor.WordVectorUtils.java

License:Apache License

public static WordVectors loadTxt(File vectorsFile, boolean... normalize) {
    AbstractCache cache = new AbstractCache<>();
    INDArray arrays[] = lines(vectorsFile.toPath()).map(line -> line.split(" "))
            .filter(fields -> fields.length > 2).map(split -> {
                VocabWord word = new VocabWord(1.0, split[0]);
                word.setIndex(cache.numWords());
                cache.addToken(word);
                cache.addWordToIndex(word.getIndex(), split[0]);
                float[] vector = new float[split.length - 1];
                range(1, split.length).parallel().forEach(i -> vector[i - 1] = parseFloat(split[i]));
                return Nd4j.create(vector);
            }).toArray(size -> new INDArray[size]);

    INDArray syn = Nd4j.vstack(arrays);/*w  ww  .  jav  a 2 s.com*/

    InMemoryLookupTable lookupTable = new InMemoryLookupTable.Builder().vectorLength(arrays[0].columns())
            .useAdaGrad(false).cache(cache).useHierarchicSoftmax(false).build();
    Nd4j.clearNans(syn);
    if (normalize.length > 0 && normalize[0]) {
        syn.diviColumnVector(syn.norm2(1));
    }

    lookupTable.setSyn0(syn);

    WordVectorsImpl vectors = new WordVectorsImpl();
    vectors.setLookupTable(lookupTable);
    vectors.setVocab(cache);
    return vectors;
}

---