Prev: MEAN AND FUCTION
Next: how to get contours into an image
From: Verena Feirer on 2 Jun 2010 08:53
Hello!


I am currently struggling with the Optimization Toolbox in combination with function handles and was hoping you might be able to help me.

I am trying to optimize a function with fminunc and have written a myfunction.m file that takes three inputs (some vectors p, n and k) and returns one value of the function myfunction. (It's some kind of p.d.f. that returns a probability for given parameters p, n and k.)
The vectors n and k have fixed/known/real values once I start the optimization. But I want to run various optimizations so I do not want to "hard-code" their values into my myfunction.m file. Instead, I want to supply them at the beginning of the specific optimization and then use a function handle on myfunction.m.
Then, I want to optimize the function in a for the fixed values of n and k. So the whole setting looks like

p = rand(10,1); % 10 vaules out of (0,1)
n = 16*ones(10,1); % a vector of 10 '16'-entries
k = binornd(n, p); % a vector of 10 (16,p_i) distributed binomials

fh = @(p)myfunction(p, n, k); % for known/defined vectors n and k
[new_p, fval, exitflag, output] = fminunc(fh, start_p, options);
% for known values of start_p

So far, so good.
BUT: My problem now is that I want to supply a gradient as well. So I add the gradient to my myfunction.m which now returns two "values" f and g, changing the function to
function [f, g] = myfunction(p, n, k)
I then tell fminunc it will recieve a gradient as well.
And that is where I fail to supply my known values n and k to the function and still persuade it to return both the function itself and its gradient, both in the variable p. When calling

options = optimset('Display', 'iter', 'GradObj','on', 'LargeScale', 'on');
fh = @(p)myfunction(p, n, k);
[new_p, fval, exitflag, output] = fminunc(fh, start_p, options);

Matlab now tells me that

??? Error using ==> uminus
Too many output arguments.

Error in ==> @(p)myfunction(p,n,k)

Error in ==> fminunc at 236
[f,GRAD(:)] = feval(funfcn{3},x,varargin{:});

Caused by: Failure in initial user-supplied objective function evaluation. FMINUNC cannot continue.

I am really at a loss here since my knowledge of working with '@'s is limited to what I found in the Optimization Toolbox manual and Matlab's help on 'function_handle'.

I hope I have provided all the important information you might need. Any help will be very appreciated.

Verena
 | 
Pages: 1
Prev: MEAN AND FUCTION
Next: how to get contours into an image