/// <summary>
/// Constructs a new Hidden Markov Model.
/// </summary>
/// <param name="symbols">The number of output symbols used for this model.</param>
/// <param name="states">The number of states for this model.</param>
/// <param name="type">The topology which should be used by this model.</param>
public HiddenMarkovModel(int symbols, int states, HiddenMarkovModelType type)
: this(null, null, states, type)
{
if (symbols <= 0)
{
throw new ArgumentOutOfRangeException("symbols",
"Number of symbols should be higher than zero.");
}
this.symbols = symbols;
this.B = new double[states, symbols];
// Initialize B with uniform probabilities
for (int i = 0; i < states; i++)
{
for (int j = 0; j < symbols; j++)
{
B[i, j] = 1.0 / symbols;
}
}
}
/// <summary>
/// Constructs a new Hidden Markov Model Sequence Classifier.
/// </summary>
public MarkovSequenceClassifier(int classes, int symbols, int[] states, String[] names, HiddenMarkovModelType type)
{
models = new HiddenMarkovModel[classes];
// For each of the possible output classes
for (int i = 0; i < classes; i++)
{
// Create a Hidden Markov Models with the given number of symbols and states
models[i] = new HiddenMarkovModel(symbols, states[i], type);
if (names != null)
{
models[i].Tag = names[i];
}
}
}
/// <summary>
/// Constructs a new Hidden Markov Model.
/// </summary>
private HiddenMarkovModel(double[,] transitions, double[] probabilities, int?states, HiddenMarkovModelType type)
{
// Automatically determine missing parameters
#region Number of states N
if (states != null)
{
if (states <= 0)
{
throw new ArgumentOutOfRangeException("states",
"Number of states should be higher than zero.");
}
}
else
{
if (probabilities != null)
{
states = probabilities.Length;
}
else if (transitions != null)
{
states = transitions.GetLength(0);
}
else
{
throw new ArgumentException("Number of states could not be determined.",
"states");
}
}
int n = states.Value;
this.states = n;
#endregion
#region Transitions Matrix A
if (transitions != null)
{
if (transitions.GetLength(0) != states)
{
throw new ArgumentException(
"Transition matrix should have the same dimensions as the number of model states.",
"transitions");
}
if (transitions.GetLength(0) != transitions.GetLength(1))
{
throw new ArgumentException("Transition matrix should be square.", "transitions");
}
}
else
{
if (type == HiddenMarkovModelType.Ergodic)
{
// Create A using uniform distribution
transitions = new double[n, n];
for (int i = 0; i < n; i++)
{
for (int j = 0; j < n; j++)
{
transitions[i, j] = 1.0 / n;
}
}
}
else
{
// Create A using uniform distribution,
// without allowing backward transitions.
transitions = new double[n, n];
for (int i = 0; i < n; i++)
{
for (int j = i; j < n; j++)
{
transitions[i, j] = 1.0 / (n - i);
}
}
}
}
this.A = transitions;
#endregion
#region Initial Probabilities pi
if (probabilities != null)
{
if (probabilities.Length != n)
{
throw new ArgumentException(
"Initial probabilities should have the same length as the number of model states.",
"probabilities");
}
}
else
{
// Create pi as left-to-right
probabilities = new double[n];
probabilities[0] = 1.0;
}
this.pi = probabilities;
#endregion
}
/// <summary>
/// Constructs a new Hidden Markov Model.
/// </summary>
private HiddenMarkovModel(double[,] transitions, double[] probabilities, int? states, HiddenMarkovModelType type)
{
// Automatically determine missing parameters
#region Number of states N
if (states != null)
{
if (states <= 0)
{
throw new ArgumentOutOfRangeException("states",
"Number of states should be higher than zero.");
}
}
else
{
if (probabilities != null)
{
states = probabilities.Length;
}
else if (transitions != null)
{
states = transitions.GetLength(0);
}
else
{
throw new ArgumentException("Number of states could not be determined.",
"states");
}
}
int n = states.Value;
this.states = n;
#endregion
#region Transitions Matrix A
if (transitions != null)
{
if (transitions.GetLength(0) != states)
throw new ArgumentException(
"Transition matrix should have the same dimensions as the number of model states.",
"transitions");
if (transitions.GetLength(0) != transitions.GetLength(1))
throw new ArgumentException("Transition matrix should be square.", "transitions");
}
else
{
if (type == HiddenMarkovModelType.Ergodic)
{
// Create A using uniform distribution
transitions = new double[n, n];
for (int i = 0; i < n; i++)
for (int j = 0; j < n; j++)
transitions[i, j] = 1.0 / n;
}
else
{
// Create A using uniform distribution,
// without allowing backward transitions.
transitions = new double[n, n];
for (int i = 0; i < n; i++)
for (int j = i; j < n; j++)
transitions[i, j] = 1.0 / (n - i);
}
}
this.A = transitions;
#endregion
#region Initial Probabilities pi
if (probabilities != null)
{
if (probabilities.Length != n)
throw new ArgumentException(
"Initial probabilities should have the same length as the number of model states.",
"probabilities");
}
else
{
// Create pi as left-to-right
probabilities = new double[n];
probabilities[0] = 1.0;
}
this.pi = probabilities;
#endregion
}
/// <summary>
/// Constructs a new Hidden Markov Model.
/// </summary>
/// <param name="symbols">The number of output symbols used for this model.</param>
/// <param name="states">The number of states for this model.</param>
/// <param name="type">The topology which should be used by this model.</param>
public HiddenMarkovModel(int symbols, int states, HiddenMarkovModelType type)
: this(null, null, states, type)
{
if (symbols <= 0)
{
throw new ArgumentOutOfRangeException("symbols",
"Number of symbols should be higher than zero.");
}
this.symbols = symbols;
this.B = new double[states, symbols];
// Initialize B with uniform probabilities
for (int i = 0; i < states; i++)
for (int j = 0; j < symbols; j++)
B[i, j] = 1.0 / symbols;
}
/// <summary>
/// Constructs a new Hidden Markov Model Sequence Classifier.
/// </summary>
public MarkovSequenceClassifier(int classes, int symbols, int[] states, String[] names, HiddenMarkovModelType type)
{
models = new HiddenMarkovModel[classes];
// For each of the possible output classes
for (int i = 0; i < classes; i++)
{
// Create a Hidden Markov Models with the given number of symbols and states
models[i] = new HiddenMarkovModel(symbols, states[i], type);
if (names != null) models[i].Tag = names[i];
}
}
