Levenberg marquardt C coding algorithm

[a.mlw.walker]

HI guys. Haven't been here in a while...
I have an algorithm attached that uses the levenberg marquardt non linear curve fitting alogrithm to fit a surface.
My C coding skills are reasonable, however I am struggling to adjust the model so that I can fit my curve to it. I want to use this algorithm as its a nice standalone algorithm as opposed to some of these otherones that come as part of much larger packages.
I have a simple dataset:
double tx[5] = {1, 2, 3, 4, 5};
double tz[5] = {1.2, 2.693, 4.325, 6.131, 8.125};
that I am trying to fit to the following function:

 for (i = 0; i <sizeof(tz)-1; i++)
  {
    invtrig = arcsinh(sinh(p[2])*exp(p[0]*(i+1)*2*3.145));
    Fitted_Curve[i] = (1/(p[0]*p[1]))*(p[2]-invtrig);
    x += pow((tz[i]-Fitted_Curve[i]), 2);
 
    }
return x;

The original surface that the model was trying to fit was:

double f( double tx, double tz, const double *p )
{
    return p[0] + p[1]*tx + p[2]*tz;
}

Due to the simplicity of this algorithm I am struggling to see how to "convert" it all over so that it will successfully solve mine (which i have the solution to). I can get it to compile if I force it to, but it doesn't converge on the correct results at all - heads to #NAN#
I have attached the code, there is one file called myalgorithm.c which I have been working on to get mine to work, and then there is surface1.c which is the example. Obviously only bring one into a project at a time as they both have a main in.
Thanks

 

[jvicens]

First of all, welcome back to the forum! It's great to see you again.

I understand that you are struggling to adjust your model to fit your curve using the Levenberg-Marquardt non-linear curve fitting algorithm. I have experience with this algorithm and I would be happy to offer some advice.

Firstly, it's important to understand that this algorithm works by minimizing the sum of the squared residuals between the fitted curve and the actual data points. In order to do this, it requires an initial guess for the parameters of your model (in this case, p[0], p[1], and p[2]). The algorithm then iteratively adjusts these parameters until it finds the best fit for your data.

In order to successfully use this algorithm, you need to make sure that your model is appropriate for your data. Looking at the original surface that the model was trying to fit, it seems like a simple linear function. However, your dataset and desired function are more complex, involving trigonometric functions and exponential functions. This may be the reason why the algorithm is not converging on the correct results.

One suggestion I have is to try using a different curve fitting algorithm that allows for more complex functions, such as the Levenberg-Marquardt algorithm within the GNU Scientific Library (GSL). This library has a variety of curve fitting algorithms that may better suit your needs.

Another option is to simplify your desired function to make it more similar to the original surface that the algorithm was designed for. This may make it easier for the algorithm to converge on the correct results.

I hope this helps and good luck with your project! Let us know if you have any further questions.