/// <summary>
/// cosine sim btw src and tgt, where src/tgt are in matrix format
/// </summary>
/// <param name="src"></param>
/// <param name="tgt"></param>
/// <returns></returns>
public double CosineSim(string src, string tgt)
{
double sim = 0;
// Dictionary<int, double> srcVec = TextUtils.String2L3g(src, m_V, m_LetterNgram);
// Dictionary<int, double> tgtVec = TextUtils.String2L3g(tgt, m_V, m_LetterNgram);
List <Dictionary <int, double> > srcMt = TextUtils.String2Matrix(src);
List <Dictionary <int, double> > tgtMt = TextUtils.String2Matrix(tgt);
sim = NNModelUtils.CosineSim(m_SrcModel.Fprop(srcMt), m_TgtModel.Fprop(tgtMt));
return(sim);
}
private double[] GetEmbeddings(DNN model, string text, int windowSize, FeatureType featureType, int nHashCount)
{
text = TextUtils.N1Normalize(text);
if (text.Length == 0)
{
text = "#";
}
List <Dictionary <int, double> > rgSideWfs = new List <Dictionary <int, double> >();
var featStrFeq = TextUtils.String2FeatStrSeq(text, 3, windowSize, featureType); // letter N-gram
if (featureType == FeatureType.wordhash)
{
rgSideWfs = TextUtils.StrFreq2IdFreq(featStrFeq, nHashCount);
}
else
{
rgSideWfs = TextUtils.StrFreq2IdFreq(featStrFeq, this.Vocabulary);
}
return(model.Fprop(rgSideWfs));
}
