MATLAB Answers

0

TroubleShoot FMinSearch with subscript assignment mismatch

Asked by Eric Diaz on 15 Jan 2016
Latest activity Commented on by Eric Diaz on 15 Jan 2016
I keep getting the following error:
Subscripted assignment dimension mismatch.
Error in fminsearch (line 190)
fv(:,1) = funfcn(x,varargin{:})
when I try to run the following optimization
objective = @(p) log(besseli(0,(((p(1)*exp(-xData/p(2))).*yData)./Variance),1)) - (((p(1)*exp(-xData/p(2))).^2) ./ (2.*Variance));
% unconstrained nonlinear optimization
parEst = fminsearch(objective,p0);
p0 is [1,2]. Variance is a predefined scalar. xData and yData are both [1,8].
Please help me as I have tried reading the help documents for fminsearch and for anonymous functions, but can't figure it out.

  0 Comments

Sign in to comment.

Tags

2 Answers

Answer by Torsten
on 15 Jan 2016
 Accepted Answer

A scalar must be returned to fminsearch from "objective" ; you return a vector.
Best wishes
Torsten.

  3 Comments

Oh, okay. Hmmm...well, that makes things difficult for me. I am very familiar with using lsqcurvefit, which does not seem to have any problem with taking vectors of data. Do you have any suggestions on how to fix the issue?
You will have to supply the sum of squared differences of your model expression and your experimental data. But I must admit, since your function involves both xtata and ydata, I don't know what is yi_model and what is yi_experiment.
Best wishes
Torsten.
Initially, I tried to break down my objective function, however I ran into issues with not having my p (parameter vector) variable defined as Matlab executed the term1 definition. So, I just put my model straight into my objective function in my previous post.
This is how I would like to write my code, as it reads more clearly. Please note that I put the sum function around the objective function below. Is the correct way to do the sum? Also, do you know a way around not having the parameter vector defined?
signalModel = p(1)*exp(-xData/p(2));
term1 = (signalModel.*yData)./Variance;
term2 = (signalModel.^2) ./ (2.*Variance);
objective = @(p) sum(log(besseli(0,term1,1)) - term2);

Sign in to comment.


Answer by John D'Errico
on 15 Jan 2016

This happens over and over again. fminsearch is an OPTIMIZER. It does not do nonlinear regression. There is a difference.
An optimizer finds the minimum value of a general function of multiple parameters. So a SCALAR valued function of one or more variables.
A nonlinear regression can be turned into an optimization by forming the sum of squares of residuals, and minimizing that as an objective.
But if you just throw a vector into a tool like fminsearch, expect it to fail. fminsearch has no idea what you want to do. Computers cannot read your mind.

  1 Comment

Thanks for your input John. I think what I need to do is just figure out the correct way to do that summation over the time points.

Sign in to comment.