private void InitNetwork()
{
long wordIndex, dimensionIndex;
ulong nextRandom = 1;
var numberOfWords = _wordCollection.GetNumberOfUniqueWords();
_hiddenLayerWeights = new float[numberOfWords, _numberOfDimensions];
if (_negative > 0)
{
_outputLayerWeights = new float[_wordCollection.GetNumberOfUniqueWords(), _numberOfDimensions];
for (wordIndex = 0; wordIndex < _wordCollection.GetNumberOfUniqueWords(); wordIndex++)
{
for (dimensionIndex = 0; dimensionIndex < _numberOfDimensions; dimensionIndex++)
{
_outputLayerWeights[wordIndex, dimensionIndex] = 0;
}
}
}
for (wordIndex = 0; wordIndex < _wordCollection.GetNumberOfUniqueWords(); wordIndex++)
{
for (dimensionIndex = 0; dimensionIndex < _numberOfDimensions; dimensionIndex++)
{
nextRandom = LinearCongruentialGenerator(nextRandom);
_hiddenLayerWeights[wordIndex, dimensionIndex] = ((nextRandom & 0xFFFF) / (float)65536 - (float)0.5) / _numberOfDimensions;
}
}
var huffmanTree = new HuffmanTree();
huffmanTree.Create(_wordCollection);
GC.Collect();
}
private void Setup()
{
_alpha = _startingAlpha;
_fileHandler.GetWordDictionaryFromFile(_wordCollection, MaxCodeLength);
// TODO: refactor the huffman stuff, so we can use it elsewhere
_wordCollection.RemoveWordsWithCountLessThanMinCount(_minCount);
var huffmanTree = new HuffmanTree();
huffmanTree.Create(_wordCollection);
InitNetwork();
if (_negativeSamples > 0)
{
InitUnigramTable();
}
GC.Collect();
}