public OneDObjectiveFunc(double[] Xobdata, double[] obdatay, double[] para)//matlab test
{
this.Xobdata = Xobdata;
this.obdatay = obdatay;
//this.noise = noise;
this.para = para;
Y = GPwithMeanfunction.GetfVactor(obdatay);
}
public OneDObjectiveFunc(double[,] MDXobdata, double[] obdatay, double[] para)
{
this.MDXobdata = MDXobdata;
this.obdatay = obdatay;
// this.noise = noise;
this.para = para;
Y = GPwithMeanfunction.GetfVactor(obdatay);
}
private static double[] GetcoveriantArray(KeyValuePair <List <double[]>, List <double[, ]> > initialclusture)
{
var obdataX = new double[initialclusture.Value.Count, initialclusture.Value[0].GetLength(1) - 1];
var obdataY = new double[initialclusture.Value.Count];
for (int i = 0; i < initialclusture.Value.Count; i++)
{
obdataY[i] = initialclusture.Value[i][0, initialclusture.Value[0].GetLength(1) - 1];
for (int j = 0; j < initialclusture.Value[0].GetLength(1) - 1; j++)
{
obdataX[i, j] = initialclusture.Value[i][0, j];
}
}
var cov = GPwithMeanfunction.GetCovMatrix(obdataX, initialclusture.Key[2], true);
return(cov.Reshape());
}
private static void GetCoveriantandInvArray(KeyValuePair <List <double[]>, List <double[, ]> > initialclusture, out double[] covarray, out double[] invarray)
{
var obdataX = new double[initialclusture.Value.Count, initialclusture.Value[0].GetLength(1) - 1];
var obdataY = new double[initialclusture.Value.Count];
for (int i = 0; i < initialclusture.Value.Count; i++)
{
obdataY[i] = initialclusture.Value[i][0, initialclusture.Value[0].GetLength(1) - 1];
for (int j = 0; j < initialclusture.Value[0].GetLength(1) - 1; j++)
{
obdataX[i, j] = initialclusture.Value[i][0, j];
}
}
var cov = GPwithMeanfunction.GetCovMatrix(obdataX, initialclusture.Key[2], true);
invarray = cov.inverse().Reshape();
covarray = cov.Reshape();
}
public static List <double> newMDGetMeanAndVar(KeyValuePair <List <double[]>, List <double[, ]> > trainedcluster, double[] testpoint)
{
var testpoints = RowtoMatrix(testpoint);
double[,] MeanStart, Xobdata;
double[] obdatay;
GetXfromValues(trainedcluster.Value, out Xobdata, out obdatay);
var GPpara = trainedcluster.Key[2];
var Regpara = trainedcluster.Key[1];
var mean = GPwithMeanfunction.GetfVactor(obdatay);
// var cov = GPwithMeanfunction.GetCovMatrix(Xobdata, GPpara, true);
var cov = trainedcluster.Key[3].Reshape(obdatay.Length, obdatay.Length, MatrixOrder.CRowMajor);
var kInv = trainedcluster.Key[4].Reshape(obdatay.Length, obdatay.Length, MatrixOrder.CRowMajor);
var MeanAndVar = new List <double>();
var a = testpoints.GetLength(0);
double variance;
var h = RowtoMatrix(CalculatRegression.gety(Xobdata, Regpara));
var hstarts = RowtoMatrix(CalculatRegression.gety(testpoints, Regpara));
int c = Xobdata.GetLength(0) - 1;
for (var i = 0; i <= a - 1; i++)
{
var XobAndTestData = GPwithMeanfunction.AddnewPoint(Xobdata, testpoints, i);
var newcov = GPwithMeanfunction.GetCovMatrix(XobAndTestData, GPpara, true);
var kstart = new double[newcov.GetLength(0) - 1, 1];
for (var j = 0; j < newcov.GetLength(0) - 1; j++)
{
kstart[j, 0] = newcov[j, newcov.GetLength(0) - 1];
}
var kstartT = kstart.transpose();
// MeanStart = (kstartT.multiply(kInv)).multiply(mean);
if (GPpara.Length - Xobdata.GetLength(1) == 2)
{
MeanStart = (kstartT.multiply(kInv)).multiply(mean);
var VarStart = (((kstartT.multiply(kInv)).multiply(kstart)).multiply(-1));
variance = VarStart[0, 0] + cov[c, c];
}
else //with mean fuc
{
var regy = CalculatRegression.gety(testpoints, Regpara);
// var regy = CalculatRegression.gety(testpoints, i, actionname)[0, 0] * GPpara[GPpara.Length - 1];
var regym = new double[1, 1];
regym[0, 0] = regy[0];
var fb = RowtoMatrix(CalculatRegression.gety(Xobdata, Regpara)).transpose().multiply(GPpara[GPpara.Length - 1]);
MeanStart = regym.add((kstartT.multiply(kInv)).multiply(mean.subtract(fb)));
//var VarStart = (((kstartT.multiply(kInv)).multiply(kstart)).multiply(-1));
//var variancestar = VarStart[0, 0] + cov[c, c];
var hstart = new double[1, 1];
hstart[0, 0] = hstarts[0, i];
var r = hstart.subtract((h.multiply(kInv)).multiply(kstart));
var hkk = ((h.multiply(kInv)).multiply(h.transpose())).inverse();
//variance = variancestar + (r.transpose().multiply(hkk))[0, 0];
variance = cov[c, c] + (r.transpose().multiply(hkk))[0, 0];
}
MeanAndVar.Add(MeanStart[0, 0]);
//
//double sdpow = Math.Pow(sd, 2);
double sdabsolute = Math.Sqrt(Math.Abs(variance));
MeanAndVar.Add(sdabsolute);
}
return(MeanAndVar);
}
public double calculate(double[] x)//matlab test
{
var matlab = true;
var md = true;
double[,] k;
if (matlab && !md)
{
k = GPwithMeanfunction.SEKernel(Xobdata, x, true);
}
else if (matlab && md)
{
k = GPwithMeanfunction.SEKernel(MDXobdata, x, true);
}
else if (x.GetLength(0) == 2)
{
k = GPwithMeanfunction.SEKernel(Xobdata, x, noise);
}
else
{
k = GPwithMeanfunction.SEKernel(MDXobdata, x, noise, true);
}
var liknoise = Math.Exp(2 * para[3]); //MD
var l = Cholesky(k.divide(liknoise).add(CreatImatrix(MDXobdata.GetLength(0)))); //MD
//var liknoise = Math.Exp(2 * para[2]);//1D
//var l = Cholesky(k.divide(liknoise).add(CreatImatrix(Xobdata.GetLength(0))));//1D
var y = l.inverse();
var l1 = y.multiply(obdatay);
l = l.transpose();
l = l.inverse();
var alpha = l.multiply(l1).divide(liknoise);
l = Cholesky(k.divide(liknoise).add(CreatImatrix(MDXobdata.GetLength(0))));//MD
// l = Cholesky(k.divide(liknoise).add(CreatImatrix(Xobdata.GetLength(0))));//1D
var f = alpha.multiply(Y);
var firstp = f[0] / 2;
var secondp = sumlogl(l);
var thrdp = l.GetLength(0) * Math.Log(2 * Math.PI * liknoise) / 2;
var total = firstp + secondp + thrdp;
//if (x.Any(a => a.CompareTo(0) == -1)//0 is sf, 1 is L
if (x[0] < 0 || x[1] < 0 ||
x[2] < 0 ||
x[3] > 1
)
{
return(10000000);
}
if (total == 100000)
{
int xx = 1;
}
return(total);
///OLD log likelihoold
//var firspart = -0.5 * VecToDouble((Y.transpose()).multiply(StarMath.inverse(k)).multiply(Y));
//var kt = StarMath.CholeskyDecomposition(k);
//var kkt = k.determinant();
//var secondpart = -0.5 * Math.Log(k.determinant());
//if (x.Any(a => a.CompareTo(0) == -1)//0 is sf, 1 is L
// //|| x[0] > 0.5
// // || x[1] < 0.1
// //|| x[2] > 1.1
// //|| x[3] > 1.1
// //|| x[2] < 0.8
// //|| x[3] < 0.8
// )
//{
// return 10000000;
//}
////else if (-(firspart + secondpart) < -1000)
////{
//// var a = -(firspart + secondpart);
//// return 10000000;
////}
//var re = -(firspart + secondpart);
// return -(firspart + secondpart);
}
