topic Is SAS IML NLPLM function equivalent to MATlab lsqcurvefit function? in SAS/IML Software and Matrix Computations

Is SAS IML NLPLM function equivalent to MATlab lsqcurvefit function?

FB1 — Mon, 28 Jan 2019 10:03:22 GMT

Hello SAS experts,

I'm trying to move a script from Matlab to SAS IML. I did it and it works apparently, but I am getting different results with the functions:

Matlab

x = lsqcurvefit(f, p, GC, DD');

SAS:

/* fun=f-dd and GC already set as parameter in fun.*/
CALL NLPLM (rc, xr, "fun", p, opt);

Are they equivalent? According to the documentation it seems so.

Thanks a lot.

Re: Is SAS IML NLPLM function equivalent to MATlab lsqcurvefit function?

Rick_SAS — Mon, 28 Jan 2019 10:15:13 GMT

You do not provide any details, so I will respond by saying that a lot of choices goes into implementing an optimization algorithm. In addition to the basic algorithm, there are finite difference derivatives (forward vs central), line search techniques, and dozens of parameters that are used to control the performance and stopping criteria. Most software provides default choices, but the defaults from one package may not match the defaults in another package. For simple problems, you might hope that the parameter estimates from two different software to agree to a few decimal places. For problems that have multiple extrema, you might converge to different local minima.