public void AKN_HMM_Learning_UT()
{
AKN_HiddenMarkovModel <String> m_HiddenMarkovModel = new AKN_HiddenMarkovModel <String>(3, 2);
m_HiddenMarkovModel.MSetStates(new float[] { 0.656F, 0.344F, 0.0F }, new string[] { "Etat_1", "Etat_2", "Etat_3" });
m_HiddenMarkovModel.MSetTransitions(new float[, ] {
{ 0.346F, 0.365F, 0.289F }, { 0.159F, 0.514F, 0.327F }, { 0.377F, 0.259F, 0.364F }
});
m_HiddenMarkovModel.MSetObservations(new String[] { "Pile", "Face" });
m_HiddenMarkovModel.MSetEmissions(new float[, ] {
{ 1.0F, 0.0F }, { 0.631F, 0.369F }, { 0.0F, 1.0F }
});
String[] observation = new String[] { "Pile", "Face", "Face", "Pile", "Pile" };
m_HiddenMarkovModel.MUpdate(observation, 0, 15);
float p = m_HiddenMarkovModel.MEvaluate(observation, eUseAlgorithme.kForward);
Assert.That(NearlyEqual(p, (float)(0.2479884)), "Problem = " + (p - 0.2479884f));
Assert.That(NearlyEqual(m_HiddenMarkovModel["Etat_1"].m_startProbability, 1.0F), "Problem !" + m_HiddenMarkovModel["Etat_1"].m_startProbability);
Assert.That(NearlyEqual(m_HiddenMarkovModel["Etat_2"].m_startProbability, 0.0F), "Problem !" + m_HiddenMarkovModel["Etat_2"].m_startProbability);
Assert.That(NearlyEqual(m_HiddenMarkovModel["Etat_3"].m_startProbability, 0.0F), "Problem !" + m_HiddenMarkovModel["Etat_3"].m_startProbability);
Assert.That(NearlyEqual(m_HiddenMarkovModel["Etat_1", "Etat_1"], 0.0F), "Problem !" + m_HiddenMarkovModel["Etat_1", "Etat_1"]);
Assert.That(NearlyEqual(m_HiddenMarkovModel["Etat_1", "Etat_2"], 0.0F), "Problem !" + m_HiddenMarkovModel["Etat_1", "Etat_2"]);
Assert.That(NearlyEqual(m_HiddenMarkovModel["Etat_1", "Etat_3"], 1.0F), "Problem !" + m_HiddenMarkovModel["Etat_1", "Etat_3"]);
Assert.That(NearlyEqual(m_HiddenMarkovModel["Etat_2", "Etat_1"], 0.2119321F), "Problem !" + m_HiddenMarkovModel["Etat_2", "Etat_1"]);
Assert.That(NearlyEqual(m_HiddenMarkovModel["Etat_2", "Etat_2"], 0.788068F), "Problem !" + m_HiddenMarkovModel["Etat_2", "Etat_2"]);
Assert.That(NearlyEqual(m_HiddenMarkovModel["Etat_2", "Etat_3"], 0.0F), "Problem !" + m_HiddenMarkovModel["Etat_2", "Etat_3"]);
Assert.That(NearlyEqual(m_HiddenMarkovModel["Etat_3", "Etat_1"], 0.0F), "Problem !" + m_HiddenMarkovModel["Etat_3", "Etat_1"]);
Assert.That(NearlyEqual(m_HiddenMarkovModel["Etat_3", "Etat_2"], 0.5147933F), "Problem !" + m_HiddenMarkovModel["Etat_3", "Etat_2"]);
Assert.That(NearlyEqual(m_HiddenMarkovModel["Etat_3", "Etat_3"], 0.4852067F), "Problem !" + m_HiddenMarkovModel["Etat_3", "Etat_3"]);
Assert.That(NearlyEqual(m_HiddenMarkovModel.MGetEmission("Etat_1", "Pile"), 1.0F), "Problem !" + m_HiddenMarkovModel.MGetEmission("Etat_1", "Pile"));
Assert.That(NearlyEqual(m_HiddenMarkovModel.MGetEmission("Etat_1", "Face"), 0.0F), "Problem !" + m_HiddenMarkovModel.MGetEmission("Etat_1", "Face"));
Assert.That(NearlyEqual(m_HiddenMarkovModel.MGetEmission("Etat_2", "Pile"), 0.9686517F), "Problem !" + m_HiddenMarkovModel.MGetEmission("Etat_2", "Pile"));
Assert.That(NearlyEqual(m_HiddenMarkovModel.MGetEmission("Etat_2", "Face"), 0.03134831F), "Problem !" + m_HiddenMarkovModel.MGetEmission("Etat_2", "Face"));
Assert.That(NearlyEqual(m_HiddenMarkovModel.MGetEmission("Etat_3", "Pile"), 0.0F), "Problem !" + m_HiddenMarkovModel.MGetEmission("Etat_3", "Pile"));
Assert.That(NearlyEqual(m_HiddenMarkovModel.MGetEmission("Etat_3", "Face"), 1.0F), "Problem !" + m_HiddenMarkovModel.MGetEmission("Etat_3", "Face"));
}
public void AKN_HMM_Decoding_UT()
{
AKN_HiddenMarkovModel <String> m_HiddenMarkovModel = new AKN_HiddenMarkovModel <String>(3, 2);
m_HiddenMarkovModel.MSetStates(new float[] { 0.656F, 0.344F, 0.0F }, new string[] { "Etat_1", "Etat_2", "Etat_3" });
m_HiddenMarkovModel.MSetTransitions(new float[, ] {
{ 0.346F, 0.365F, 0.289F }, { 0.159F, 0.514F, 0.327F }, { 0.377F, 0.259F, 0.364F }
});
m_HiddenMarkovModel.MSetObservations(new String[] { "Pile", "Face" });
m_HiddenMarkovModel.MSetEmissions(new float[, ] {
{ 1.0F, 0.0F }, { 0.631F, 0.369F }, { 0.0F, 1.0F }
});
String[] observation = new String[] { "Pile", "Face", "Face", "Pile", "Pile" };
int[] Path = m_HiddenMarkovModel.MDecode(observation);
Assert.That(Path.Length == 5, "Problem ! " + Path.Length);
Assert.That(Path[0] == 0, "Problem ! " + Path[0]);
Assert.That(Path[1] == 2, "Problem ! " + Path[1]);
Assert.That(Path[2] == 2, "Problem ! " + Path[2]);
Assert.That(Path[3] == 0, "Problem ! " + Path[3]);
Assert.That(Path[4] == 0, "Problem ! " + Path[4]);
}
public void AKN_HMM_Creation_UT()
{
//
// ┌───────┐ ┌────────┐
// 0,2 │ Sport │ │ Sport │ 0,5
// 0,1 │ Marche│ │ Marche │ 0,3
// 0,7 │ Etude │ 0,4 │ Etude │ 0,2
// ┌┴───────┴┐<─────────────┌┴────────┴┐
// ┌───│ Pluie │ │ Soleil │───┐
// 0,7 └──>└─────────┘─────────────>└──────────┘<──┘ 0,6
// ▲ 0,3 ▲
// │ │
// │ │
// 0,6 0,4
//
//Création d'un modéle de markov caché avec deux états cachées et 3 observation
AKN_HiddenMarkovModel <String> m_HiddenMarkovModel_1 = new AKN_HiddenMarkovModel <String>(2, 3);
//Affectation des probabilité de démarrage aux états
//Les noms des états sont facultatifs
m_HiddenMarkovModel_1.MSetStates(new float[] { 0.6F, 0.4F }, new string[] { "Pluie", "Soleil" });
//Affectation des probabilités aux transitions
//Une transition non affecté vaut zéro.
m_HiddenMarkovModel_1["Pluie", "Pluie"] = 0.7F;
m_HiddenMarkovModel_1["Pluie", "Soleil"] = 0.3F;
m_HiddenMarkovModel_1["Soleil", "Pluie"] = 0.4F;
m_HiddenMarkovModel_1["Soleil", "Soleil"] = 0.6F;
// Affection des nom aux états observables
// Cette étape et facultatifs
m_HiddenMarkovModel_1.MSetObservations(new String[] { "Sport", "Marche", "Etude" });
//Affectation des probabilités des états observables
// 1(Cette options necessite l'affectation des noms aux observations et aux états)
m_HiddenMarkovModel_1.MSetEmission("Pluie", "Sport", 0.2F);
m_HiddenMarkovModel_1.MSetEmission("Pluie", "Marche", 0.1F);
m_HiddenMarkovModel_1.MSetEmission("Pluie", "Etude", 0.7F);
m_HiddenMarkovModel_1.MSetEmission("Soleil", "Sport", 0.5F);
m_HiddenMarkovModel_1.MSetEmission("Soleil", "Marche", 0.3F);
m_HiddenMarkovModel_1.MSetEmission("Soleil", "Etude", 0.2F);
// 2
m_HiddenMarkovModel_1.MSetEmissionByIndex(0, 0, 0.2F);
m_HiddenMarkovModel_1.MSetEmissionByIndex(0, 1, 0.1F);
m_HiddenMarkovModel_1.MSetEmissionByIndex(0, 2, 0.7F);
m_HiddenMarkovModel_1.MSetEmissionByIndex(1, 0, 0.5F);
m_HiddenMarkovModel_1.MSetEmissionByIndex(1, 1, 0.3F);
m_HiddenMarkovModel_1.MSetEmissionByIndex(1, 2, 0.2F);
Assert.That(m_HiddenMarkovModel_1.MGetEmissionByIndex(0, 0) == 0.2F, "Problem !");
Assert.That(m_HiddenMarkovModel_1.MGetEmissionByIndex(0, 1) == 0.1F, "Problem !");
Assert.That(m_HiddenMarkovModel_1.MGetEmissionByIndex(0, 2) == 0.7F, "Problem !");
Assert.That(m_HiddenMarkovModel_1.MGetEmissionByIndex(1, 0) == 0.5F, "Problem !");
Assert.That(m_HiddenMarkovModel_1.MGetEmissionByIndex(1, 1) == 0.3F, "Problem !");
Assert.That(m_HiddenMarkovModel_1.MGetEmissionByIndex(1, 2) == 0.2F, "Problem !");
Assert.That(m_HiddenMarkovModel_1.MGetEmission("Pluie", "Sport") == 0.2F, "Problem !");
Assert.That(m_HiddenMarkovModel_1.MGetEmission("Pluie", "Marche") == 0.1F, "Problem !");
Assert.That(m_HiddenMarkovModel_1.MGetEmission("Pluie", "Etude") == 0.7F, "Problem !");
Assert.That(m_HiddenMarkovModel_1.MGetEmission("Soleil", "Sport") == 0.5F, "Problem !");
Assert.That(m_HiddenMarkovModel_1.MGetEmission("Soleil", "Marche") == 0.3F, "Problem !");
Assert.That(m_HiddenMarkovModel_1.MGetEmission("Soleil", "Etude") == 0.2F, "Problem !");
AKN_HiddenMarkovModel <String> m_HiddenMarkovModel_2 = new AKN_HiddenMarkovModel <String>(2, 2);
m_HiddenMarkovModel_2.MSetStates(new float[] { 1.0F, 0.0F }, new string[] { "Samedi", "Dimanche" });
m_HiddenMarkovModel_2.MSetTransitions(new float[, ] {
{ 0.5F, 0.5F }, { 0.0F, 1.0F }
});
m_HiddenMarkovModel_2.MSetObservations(new String[] { "Soleil", "Pluie" });
m_HiddenMarkovModel_2.MSetEmissions(new float[, ] {
{ 0.3F, 0.7F }, { 0.8F, 0.2F }
});
Assert.That(m_HiddenMarkovModel_2["Samedi"].m_startProbability == 1.0F, "Problem !");
Assert.That(m_HiddenMarkovModel_2["Dimanche"].m_startProbability == 0.0F, "Problem !");
Assert.That(m_HiddenMarkovModel_2["Samedi", "Samedi"] == 0.5F, "Problem !");
Assert.That(m_HiddenMarkovModel_2["Samedi", "Dimanche"] == 0.5F, "Problem !");
Assert.That(m_HiddenMarkovModel_2["Dimanche", "Samedi"] == 0.0F, "Problem !");
Assert.That(m_HiddenMarkovModel_2["Dimanche", "Dimanche"] == 1.0F, "Problem !");
Assert.That(m_HiddenMarkovModel_2.MGetEmission("Samedi", "Soleil") == 0.3F, "Problem !");
Assert.That(m_HiddenMarkovModel_2.MGetEmission("Samedi", "Pluie") == 0.7F, "Problem !");
Assert.That(m_HiddenMarkovModel_2.MGetEmission("Dimanche", "Soleil") == 0.8F, "Problem !");
Assert.That(m_HiddenMarkovModel_2.MGetEmission("Dimanche", "Pluie") == 0.2F, "Problem !");
}
public void AKN_HMM_Evaluation_UT()
{
AKN_HiddenMarkovModel <String> m_HiddenMarkovModel;
String[] observations;
float result;
// Exemple 1
m_HiddenMarkovModel = new AKN_HiddenMarkovModel <String>(2, 2);
m_HiddenMarkovModel.MSetStates(new float[] { 1.0F, 0.0F }, new string[] { "Samedi", "Dimanche" });
m_HiddenMarkovModel.MSetTransitions(new float[, ] {
{ 0.5F, 0.5F }, { 0.0F, 1.0F }
});
m_HiddenMarkovModel.MSetObservations(new String[] { "Soleil", "Pluie" });
m_HiddenMarkovModel.MSetEmissions(new float[, ] {
{ 0.3F, 0.7F }, { 0.8F, 0.2F }
});
observations = new String[] { "Soleil", "Pluie", "Pluie", "Soleil" };
result = m_HiddenMarkovModel.MEvaluate(observations, eUseAlgorithme.kForward);
Assert.That(NearlyEqual(result, 0.0334125F), "Problem ! result = " + result);
result = m_HiddenMarkovModel.MEvaluate(observations, eUseAlgorithme.kBackward);
Assert.That(NearlyEqual(result, 0.0334125F), "Problem ! result = " + result);
result = m_HiddenMarkovModel.MEvaluate(observations, eUseAlgorithme.kForwardAndBackward, 0);
Assert.That(NearlyEqual(result, 0.0334125F), "Problem ! result = " + result);
result = m_HiddenMarkovModel.MEvaluate(observations, eUseAlgorithme.kForwardAndBackward, 1);
Assert.That(NearlyEqual(result, 0.0334125F), "Problem ! result = " + result);
result = m_HiddenMarkovModel.MEvaluate(observations, eUseAlgorithme.kForwardAndBackward, 2);
Assert.That(NearlyEqual(result, 0.0334125F), "Problem ! result = " + result);
result = m_HiddenMarkovModel.MEvaluate(observations, eUseAlgorithme.kForwardAndBackward, 3);
Assert.That(NearlyEqual(result, 0.0334125F), "Problem ! result = " + result);
// Exemple 2
m_HiddenMarkovModel = new AKN_HiddenMarkovModel <String>(3, 2);
m_HiddenMarkovModel.MSetStates(new float[] { 0.5F, 0.3F, 0.2F }, new string[] { "Etat_1", "Etat_2", "Etat_3" });
m_HiddenMarkovModel.MSetTransitions(new float[, ] {
{ 0.45F, 0.35F, 0.2F }, { 0.1F, 0.5F, 0.4F }, { 0.15F, 0.25F, 0.6F }
});
m_HiddenMarkovModel.MSetObservations(new String[] { "Pile", "Face" });
m_HiddenMarkovModel.MSetEmissions(new float[, ] {
{ 1.0F, 0.0F }, { 0.5F, 0.5F }, { 0.0F, 1.0F }
});
observations = new String[] { "Pile", "Face", "Face", "Pile", "Pile" };
result = m_HiddenMarkovModel.MEvaluate(observations, eUseAlgorithme.kForward);
Assert.That(NearlyEqual(result, 0.0277625F), "Problem ! result = " + result);
result = m_HiddenMarkovModel.MEvaluate(observations, eUseAlgorithme.kBackward);
Assert.That(NearlyEqual(result, 0.0277625F), "Problem ! result = " + result);
result = m_HiddenMarkovModel.MEvaluate(observations, eUseAlgorithme.kForwardAndBackward, 0);
Assert.That(NearlyEqual(result, 0.0277625F), "Problem ! result = " + result);
result = m_HiddenMarkovModel.MEvaluate(observations, eUseAlgorithme.kForwardAndBackward, 1);
Assert.That(NearlyEqual(result, 0.0277625F), "Problem ! result = " + result);
result = m_HiddenMarkovModel.MEvaluate(observations, eUseAlgorithme.kForwardAndBackward, 2);
Assert.That(NearlyEqual(result, 0.0277625F), "Problem ! result = " + result);
result = m_HiddenMarkovModel.MEvaluate(observations, eUseAlgorithme.kForwardAndBackward, 3);
Assert.That(NearlyEqual(result, 0.0277625F), "Problem ! result = " + result);
result = m_HiddenMarkovModel.MEvaluate(observations, eUseAlgorithme.kForwardAndBackward, 4);
Assert.That(NearlyEqual(result, 0.0277625F), "Problem ! result = " + result);
}