private VectorizedDictionary Train(Corpus.Corpus corpus, bool cBow)
{
var parameter = new WordToVecParameter();
parameter.SetCbow(cBow);
var neuralNetwork = new NeuralNetwork(corpus, parameter);
return(neuralNetwork.Train());
}
/**
* <summary>Constructor for the {@link NeuralNetwork} class. Gets corpus and network parameters as input and sets the
* corresponding parameters first. After that, initializes the network with random weights between -0.5 and 0.5.
* Constructs vector update matrix and prepares the exp table.</summary>
* <param name="corpus">Corpus used to train word vectors using Word2Vec algorithm.</param>
* <param name="parameter">Parameters of the Word2Vec algorithm.</param>
*/
public NeuralNetwork(Corpus.Corpus corpus, WordToVecParameter parameter)
{
this._vocabulary = new Vocabulary(corpus);
this._parameter = parameter;
this._corpus = corpus;
_wordVectors = new Matrix(_vocabulary.Size(), parameter.GetLayerSize(), -0.5, 0.5, new Random());
_wordVectorUpdate = new Matrix(_vocabulary.Size(), parameter.GetLayerSize());
PrepareExpTable();
}
public Corpus.Corpus Convert(IConstituencyToDependencyTreeConverter constituencyToDependencyTreeConverter)
{
var annotatedCorpus = new Corpus.Corpus();
for (var i = 0; i < _treeBank.Size(); i++)
{
annotatedCorpus.AddSentence(constituencyToDependencyTreeConverter.Convert(_treeBank.Get(i), null));
}
return(annotatedCorpus);
}
/**
* <summary>Constructor for the {@link Vocabulary} class. For each distinct word in the corpus, a {@link VocabularyWord}
* instance is created. After that, words are sorted according to their occurrences. Unigram table is constructed,
* where after Huffman tree is created based on the number of occurrences of the words.</summary>
* <param name="corpus">Corpus used to train word vectors using Word2Vec algorithm.</param>
*/
public Vocabulary(Corpus.Corpus corpus)
{
var wordList = corpus.GetWordList();
_vocabulary = new List <VocabularyWord>();
foreach (var word in wordList)
{
_vocabulary.Add(new VocabularyWord(word.GetName(), corpus.GetCount(word)));
}
_vocabulary.Sort();
CreateUniGramTable();
ConstructHuffmanTree();
_vocabulary.Sort(new TurkishWordComparator());
}
public void Setup()
{
english = new Corpus.Corpus("../../../english-similarity-dataset.txt");
turkish = new Corpus.Corpus("../../../turkish-similarity-dataset.txt");
}
public void Setup()
{
corpus = new Corpus.Corpus("../../../corpus.txt");
simpleCorpus = new Corpus.Corpus("../../../simplecorpus.txt");
}
