public static void InitialParameter(int _m, int _n, int _k)
{
m = _m + 1;
n = _n;
k = _k;
alpha = Enumerable.Range(0, m - 1).Select(dp => 1.0).ToArray();
beta = Enumerable.Range(0, k).Select(dp => 1.0).ToArray();
MathNet.Numerics.Distributions.Dirichlet dPhi = new MathNet.Numerics.Distributions.Dirichlet(alpha);
MathNet.Numerics.Distributions.Dirichlet dTheta = new MathNet.Numerics.Distributions.Dirichlet(beta);
phi = new MathNet.Numerics.Distributions.Categorical[m];
zArr = new int[n];
var p = new double[k];
for (int i = 0; i < k; i++)
p[i] = 1.0 / k;
theta = new MathNet.Numerics.Distributions.Categorical(
p
);
for (int i = 0; i < n; i++)
{
zArr[i] = theta.Sample();
}
for (int i = 0; i < m; i++)
{
phi[i] = new MathNet.Numerics.Distributions.Categorical(
dPhi.Sample()
);
}
}
/// <summary>
/// The probability density function for a primitive distribution and point.
/// Throws NotImplementedException if no PDF is defined for given distribution.
/// </summary>
public static Prob Pdf(PrimitiveDist <double[]> dist, double[] x)
{
if (dist is DirichletPrimitive)
{
var dir = dist as DirichletPrimitive;
var d = new MathNet.Numerics.Distributions.Dirichlet(dir.alpha);
return(Prob(d.Density(x)));
}
if (dist is MultiVariateNormalPrimitive)
{
var d = dist as MultiVariateNormalPrimitive;
return(Prob(d.mvn.Density(DenseMatrix.OfRowArrays(x))));
}
throw new NotImplementedException("No PDF for this distribution implemented");
}
public static void GibbsSamplingEach(List<SequenceEle> seq, int[][] matrix)
{
for (int i = 0; i < n; i++)
{
//draw z
double[] p = new double[k];
for (int z = 0; z < k; z++)
{
var theta_tmp = theta.P[z];
var phi_tmp = 0.0;
if (i == 0)
try
{
phi_tmp = phi[m - 1].P[seq[i].tIndex];
}
catch (Exception e)
{
}
else
try
{
phi_tmp = phi[seq[i - 1].tIndex].P[seq[i].tIndex];
}
catch (Exception e)
{
}
if (phi_tmp <= 0)
phi_tmp = 1.0 / Math.Pow(10, 100);
try
{
var f = Math.Exp(
(-1) * Math.Pow((z + 1 - matrix[i][seq[i].tIndex]
), 2)
);
p[z] = Math.Pow(10, 10) * theta_tmp * phi_tmp * f;
}
catch (Exception e)
{
}
}
try
{
MathNet.Numerics.Distributions.Categorical dis = new MathNet.Numerics.Distributions.Categorical(p);
zArr[i] = dis.Sample();
}
catch (Exception e)
{
}
var beta_tmp = new double[k];
beta.CopyTo(beta_tmp, 0);
foreach (var g in zArr.GroupBy(dp => dp))
{
beta_tmp[g.Key] += g.Count();
}
var d_tmp = new MathNet.Numerics.Distributions.Dirichlet(beta_tmp);
theta = new MathNet.Numerics.Distributions.Categorical(d_tmp.Sample());
for (int j = 0; j < m; j++)
{
p = Enumerable.Range(0, m).Select(dp => 1.0).ToArray();
for (int ii = 0; ii < n; ii++)
{
if (j == m - 1 && ii == 0)
{
try
{
var z = zArr[ii];
var f = Math.Exp(
(-1) * Math.Pow((z - matrix[ii][seq[ii].tIndex] / ((double)matrix[ii].Max()) * k), 2)
);
p[seq[ii].tIndex] *= phi[j].P[seq[ii].tIndex] * f;
}
catch (Exception e)
{
}
}
else
{
if (ii == 0)
continue;
try
{
if (seq[ii - 1].tIndex == j)
{
var z = zArr[ii];
var f = Math.Exp(
(-1) * Math.Pow((z + 1 - matrix[ii][seq[ii].tIndex]
), 2)
);
p[seq[ii].tIndex] *= phi[j].P[seq[ii].tIndex] * f;
}
}
catch (Exception e)
{
}
}
}
for (int s = 0; s < p.Length; s++)
if (p[s] == 1.0 || p[s] <= 0)
p[s] = 1.0 / Math.Pow(10, 100);
phi[j] = new MathNet.Numerics.Distributions.Categorical(p);
}
}
}
public DirichletPrimitive(double[] alpha, Random gen)
{
this.alpha = alpha;
dist = new MathNet.Numerics.Distributions.Dirichlet(alpha, gen);
}