static public double solve(List <Vectors> x, double[] a, ModelBase m)
{
int npts = x.Count;
int n = a.GetLength(0);
m.Normalise(a); // normalise P(0)
double e0 = Objective(x, a, m);
double lambda = lambda1;
bool done = false;
// g = gradient,H =hessian,d= step to min
// H d = -g solve for d
double [][] H = new double [n][];
//setup H
for (int i = 0; i < n; i++)
{
H[i] = new double[n];
}
//
double[] g = new double[n];
int iter = 0;
int term = 0;
do
{
++iter;
// hessian approximation
for (int r = 0; r < n; r++)
{
for (int c = 0; c < n; c++)
{
for (int i = 0; i < npts; i++)
{
if (i == 0)
{
H[r][c] = 0.0d;
}
H[r][c] += m.grad(x[i], a, r) * m.grad(x[i], a, c);
} // for all points
}
}
//set diagonal elements
for (int r = 0; r < n; r++)
{
H[r][r] *= (1 + lambda);
H[r][r] += lambda;
}
//gradient
for (int r = 0; r < n; r++)
{
for (int i = 0; i < npts; i++)
{
if (i == 0)
{
g[r] = 0.0d;
}
g[r] = g[r] - m.DistLMA(x[i], a) * m.grad(x[i], a, r);
} // for all points
}
Choleski ch = new Choleski();
double[] d = new double[a.GetLength(0)];
ch.solve(H, d, g);
double[] na = new double[n];
for (int i = 0; i < n; i++)
{
na[i] = a[i] + d[i];
}
m.Normalise(na);
double e1 = Objective(x, na, m);
if (Math.Abs(e1 - e0) > termepsilon)
{
term = 0;
}
else
{
term++;
if (termepsilon == 4)
{
done = true;
}
}
if (iter >= maxiter)
{
done = true;
}
if (e1 > e0)
{
lambda *= 10.0d;
}
else
{
lambda *= 0.04d;
e0 = e1;
for (int i = 0; i < n; i++)
{
a[i] = na[i];
}
}
}while (!done);
return(lambda);
}
private string CreateDeleteSql(ModelBase model, int id)
{
return(string.Format($"DELETE FROM {model.TableName} WHERE Id = {id}"));
}
public void UpLoad(ModelBase modelBase)
{
ModelDic.Remove(modelBase.ModelName);
}
public void Register(ModelBase modelBase)
{
ModelDic[modelBase.ModelName] = modelBase.GetType();
}
