Examples of be.ac.ulg.montefiore.run.jahmm.toolbox.KullbackLeiblerDistanceCalculator

be.ac.ulg.montefiore.run.jahmm.toolbox.KullbackLeiblerDistanceCalculator

Computes the distance between HMMs.

The distance metric is similar to the Kullback-Leibler distance defined on distributions. More information can be found in A Probabilistic Distance Measure For HMMs by Rabiner and Juang (AT&T Technical Journal, vol. 64, Feb. 1985, pages 391-408).

This distance measure is not symetric: distance(hmm1, hmm2) is not necessary equal to distance(hmm2, hmm1). To get a symetric distance definition, compute (distance(hmm1, hmm2) + distance(hmm2, hmm1)) / 2.

    
    sequences = new ArrayList<List<ObservationInteger>>();    
    for (int i = 0; i < 100; i++)
      sequences.add(mg.observationSequence(100));
    
    klc = new KullbackLeiblerDistanceCalculator();
  }

    BaumWelchLearner bwl = new BaumWelchLearner();
    
    Hmm<ObservationDiscrete<Packet>> learntHmm = buildInitHmm();
    
    // This object measures the distance between two HMMs
    KullbackLeiblerDistanceCalculator klc = 
      new KullbackLeiblerDistanceCalculator();
    
    // Incrementally improve the solution
    for (int i = 0; i < 10; i++) {
      System.out.println("Distance at iteration " + i + ": " +
          klc.distance(learntHmm, hmm));
      learntHmm = bwl.iterate(learntHmm, sequences);
    }
    
    System.out.println("Resulting HMM:\n" + learntHmm);

  throws IOException, FileFormatException
  {
    Hmm<O> hmm1 = HmmReader.read(reader1, relatedObjs.opdfReader());
    Hmm<O> hmm2 = HmmReader.read(reader2, relatedObjs.opdfReader());
    
    KullbackLeiblerDistanceCalculator kl = 
      new KullbackLeiblerDistanceCalculator();
    System.out.println(kl.distance(hmm1, hmm2));
  }

Examples of be.ac.ulg.montefiore.run.jahmm.toolbox.KullbackLeiblerDistanceCalculator

Related Classes of be.ac.ulg.montefiore.run.jahmm.toolbox.KullbackLeiblerDistanceCalculator