public void Run_DefaultModelAndObservagtionsAndNormalized_TrainedMode()
{
var startDistribution = new[] { 0.85, 0.15 };
var tpm = new double[2][];
tpm[0] = new[] { 0.3, 0.7 };
tpm[1] = new[] { 0.1, 0.9 };
var observations = new List <IObservation>
{
new Observation(new double[] { 0 }, "A"),
new Observation(new double[] { 1 }, "B"),
new Observation(new double[] { 1 }, "B"),
new Observation(new double[] { 0 }, "A")
};
var emissions = new DiscreteDistribution[2];
emissions[0] = new DiscreteDistribution(new double[] { 0, 1 }, new[] { 0.4, 0.6 });
emissions[1] = new DiscreteDistribution(new double[] { 0, 1 }, new[] { 0.5, 0.5 });
var symbols = new List <IObservation> {
new Observation(new double[] { 0 }, "A"), new Observation(new double[] { 1 }, "B")
};
var model = HiddenMarkovModelFactory.GetModel(new ModelCreationParameters <DiscreteDistribution>()
{
Pi = startDistribution, TransitionProbabilityMatrix = tpm, Emissions = emissions
}); //new HiddenMarkovModel(startDistribution, tpm, emissions) { LogNormalized = true};
model.Normalized = true;
var algo = new BaumWelch(observations, model, symbols);
var res = algo.Run(100, LikelihoodTolerance);
Assert.AreEqual(0.82585, Math.Round(res.Pi[0], 5));
Assert.AreEqual(0.17415, Math.Round(res.Pi[1], 5));
Assert.AreEqual(0.33005, Math.Round(res.TransitionProbabilityMatrix[0][0], 5));
Assert.AreEqual(0.66995, Math.Round(res.TransitionProbabilityMatrix[0][1], 5));
Assert.AreEqual(0.09871, Math.Round(res.TransitionProbabilityMatrix[1][0], 5));
Assert.AreEqual(0.90129, Math.Round(res.TransitionProbabilityMatrix[1][1], 5));
Assert.AreEqual(0.65845, Math.Round(res.Emission[0].ProbabilityMassFunction(0), 5));
Assert.AreEqual(0.34155, Math.Round(res.Emission[0].ProbabilityMassFunction(1), 5));
Assert.AreEqual(0.41225, Math.Round(res.Emission[1].ProbabilityMassFunction(0), 5));
Assert.AreEqual(0.58775, Math.Round(res.Emission[1].ProbabilityMassFunction(1), 5));
Assert.AreEqual(-2.5638779209981162, res.Likelihood);
Assert.AreEqual(1d, Math.Round(res.Pi.Sum(), 5));
Assert.AreEqual(1d, Math.Round(res.TransitionProbabilityMatrix[0].Sum(), 5));
Assert.AreEqual(1d, Math.Round(res.TransitionProbabilityMatrix[1].Sum(), 5));
Assert.AreEqual(1d, Math.Round(res.Emission[0].ProbabilityMassFunction(0) + res.Emission[0].ProbabilityMassFunction(1), 5));
Assert.AreEqual(1d, Math.Round(res.Emission[1].ProbabilityMassFunction(0) + res.Emission[1].ProbabilityMassFunction(1), 5));
}
public void Run_DefaultModelAndObservagtions_TrainedModel()
{
var startDistribution = new[] { 0.85, 0.15 };
// s = 0, t = 1
var tpm = new double[2][];
tpm[0] = new[] { 0.3, 0.7 };
tpm[1] = new[] { 0.1, 0.9 };
var observations = new List <IObservation>
{
new Observation(new double[] { 0 }, "A"),
new Observation(new double[] { 1 }, "B"),
new Observation(new double[] { 1 }, "B"),
new Observation(new double[] { 0 }, "A")
};
var emissions = new DiscreteDistribution[2];
emissions[0] = new DiscreteDistribution(new double[] { 0, 1 }, new[] { 0.4, 0.6 });
emissions[1] = new DiscreteDistribution(new double[] { 0, 1 }, new[] { 0.5, 0.5 });
var symbols = new List <IObservation> {
new Observation(new double[] { 0 }, "A"), new Observation(new double[] { 1 }, "B")
};
var model = HiddenMarkovModelFactory.GetModel(new ModelCreationParameters <DiscreteDistribution>()
{
Pi = startDistribution, TransitionProbabilityMatrix = tpm, Emissions = emissions
}); //new HiddenMarkovModel(startDistribution, tpm, emissions) { LogNormalized = false };
model.Normalized = false;
var algo = new BaumWelch(observations, model, symbols);
var res = algo.Run(100, LikelihoodTolerance);
Assert.AreEqual(0.8258482510939813, res.Pi[0]);
Assert.AreEqual(0.17415174890601867, res.Pi[1]);
Assert.AreEqual(0.330050127737348, res.TransitionProbabilityMatrix[0][0]);
Assert.AreEqual(0.669949872262652, res.TransitionProbabilityMatrix[0][1]);
Assert.AreEqual(0.098712289730350428, res.TransitionProbabilityMatrix[1][0]);
Assert.AreEqual(0.90128771026964949, res.TransitionProbabilityMatrix[1][1]);
Assert.AreEqual(0.65845050146912831, res.Emission[0].ProbabilityMassFunction(0));
Assert.AreEqual(0.34154949853087169, res.Emission[0].ProbabilityMassFunction(1));
Assert.AreEqual(0.4122484947955643, res.Emission[1].ProbabilityMassFunction(0));
Assert.AreEqual(0.58775150520443575, res.Emission[1].ProbabilityMassFunction(1));
Assert.AreEqual(0.0770055392812707, res.Likelihood);
Assert.AreEqual(1d, res.Pi.Sum());
Assert.AreEqual(1d, res.TransitionProbabilityMatrix[0].Sum());
Assert.AreEqual(1d, Math.Round(res.TransitionProbabilityMatrix[1].Sum(), 5));
Assert.AreEqual(1d, res.Emission[0].ProbabilityMassFunction(0) + res.Emission[0].ProbabilityMassFunction(1));
Assert.AreEqual(1d, res.Emission[1].ProbabilityMassFunction(0) + res.Emission[1].ProbabilityMassFunction(1));
}
public void Train(double[] observations)
{
if (_pi == null || _transitionProbabilityMatrix == null || _emission == null)
{
throw new ApplicationException("Initialize the model with initial valuesss");
}
var model = HiddenMarkovModelStateFactory.GetState(new ModelCreationParameters <DiscreteDistribution> {
Pi = _pi, TransitionProbabilityMatrix = _transitionProbabilityMatrix, Emissions = _emission
}); //new HiddenMarkovModelState<DiscreteDistribution>(_pi, _transitionProbabilityMatrix, _emission);
model.Normalized = Normalized;
var alg = new BaumWelch(Helper.Convert(observations), model, Helper.Convert(observations));
var estimatedParameters = alg.Run(100, 20);
_pi = estimatedParameters.Pi;
_transitionProbabilityMatrix = estimatedParameters.TransitionProbabilityMatrix;
_emission = estimatedParameters.Emission;
Likelihood = estimatedParameters.Likelihood;
}