Reputation: 381
I try to use the example from EJML wiki where we use Levenberg Marquardt optimization source code here
I compare it with this one .Net version in which we have the possibility to give the function model parameters.
For example: a*x^2+b*x+c
and we can give as inputs all parameters for the model.
However, for the EJML LM code, I cannot see where can I give those model parameters.
I paste below how I use the LM EJML class:
public class Main {
public static void main(String[] args) {
LevenbergMarquardt lm = new LevenbergMarquardt(new LevenbergMarquardt.Function() {
@Override
public void compute(DenseMatrix64F param, DenseMatrix64F x, DenseMatrix64F y) {
// TODO Auto-generated method stub
System.out.println("param:");
param.print();
System.out.println("X:");
x.print();
//y=a*x^2+b*x+c
for (int i = 0; i < x.numRows; i++) {
double xx = x.get(i, 0);
y.set(i, 0, param.get(0, 0) * xx * xx +
param.get(1, 0) * xx + param.get(2, 0));
}
System.out.println("Y:");
y.print();
}
});
//Seed inital parameters
lm.optimize(new DenseMatrix64F(new double[][]{{1}, {1}, {1}}),
new DenseMatrix64F(new double[][]{{0.1975}, {0.5084}, {0.7353}, {0.9706},
{1.1891}}), new DenseMatrix64F(new double[][]{{-0.0126}, {0.2311},
{0.4412}, {1.0210}, {1.1891}}));
}
}
So How can I give those model parameters?
Upvotes: 0
Views: 210
Reputation: 2812
You might want to checkout DDogleg . It contains a more robust LM solver and uses EJML under the hood. The LM solver you found on EJML's website is intended as an example for how to use EJML and skips a few details.
Upvotes: 0