public void TestDiscBWL()
{
var hmm = BuildDiscHMM();
var learntHmm = BuildDiscInitHMM();
var mg = new MarkovGenerator(hmm);
var training = mg.GenerateSequences(200, 100);
var bwl = new TrainBaumWelch(learntHmm, training);
var klc = new KullbackLeiblerDistanceCalculator();
bwl.Iteration(5);
learntHmm = (HiddenMarkovModel) bwl.Method;
var e = klc.Distance(learntHmm, hmm);
Assert.IsTrue(e < 0.01);
}
public void TestDiscKMeans()
{
var hmm = BuildDiscHMM();
var mg = new MarkovGenerator(hmm);
var sequences = mg.GenerateSequences(200, 100);
var trainer = new TrainKMeans(hmm, sequences);
var klc =
new KullbackLeiblerDistanceCalculator();
trainer.Iteration(5);
var learntHmm = (HiddenMarkovModel) trainer.Method;
var e = klc.Distance(learntHmm, hmm);
Assert.IsTrue(e < 0.05);
}
public double Distance(HiddenMarkovModel hmm1,
HiddenMarkovModel hmm2)
{
double distance = 0.0;
for (int i = 0; i < SequenceCount; i++)
{
IMLDataSet oseq = new MarkovGenerator(hmm1)
.ObservationSequence(Len);
distance += (new ForwardBackwardScaledCalculator(oseq, hmm1)
.LnProbability() - new ForwardBackwardScaledCalculator(
oseq, hmm2).LnProbability())
/ Len;
}
return(distance / SequenceCount);
}
public double Distance(HiddenMarkovModel hmm1,
HiddenMarkovModel hmm2)
{
double distance = 0.0;
for (int i = 0; i < SequenceCount; i++)
{
IMLDataSet oseq = new MarkovGenerator(hmm1)
.ObservationSequence(Len);
distance += (new ForwardBackwardScaledCalculator(oseq, hmm1)
.LnProbability() - new ForwardBackwardScaledCalculator(
oseq, hmm2).LnProbability())
/Len;
}
return distance/SequenceCount;
}
