How to see the current value of the variable in "fminsearch" optimization?

Hi,

I am running fminsearch command to find a minimum of a function. It is taking very long. However, I want to see the current value of the variable being serached. For example,"X=fminsearch(@fcn,init0,options)" showld display X in each interation. I looked into documentaiton of "optimset" but I couldn't find something helpful.

Advance, Thanks,

Ahmed.

Answer 1

[Alan Weiss]

You can do this using an Output Function. That example shows how to return the history of points, which is I think what you want.

Alan Weiss

MATLAB mathematical toolbox documentation

[Ahmed Rehan]

Thank you so much. This does the job. Based on this solution, I also tried disp() command in my function code, which also did the work.

Answer 2

[Vladimir Sovkov]

optimset('Display','iter');

From Matlab documentation:

'Display' — Level of display

'notify' (default) | 'final' | 'off' | 'none' | 'iter'

Level of display, specified as the comma-separated pair consisting of 'Display' and one of these values:

• 'notify' — Display output only if the function does not converge.
• 'final' — Display just the final output.
• 'off' or 'none' — Display no output.
• 'iter' — Display output at each iteration (not available for lsqnonneg).

Display is available for all optimization solvers.

Example: options = optimset('Display','iter')

Data Types: char | string

[Ahmed Rehan]

Thank you for the answer, but this command only displays the iteration number, not the current value of variable tht is being searched. I want the current value of X in f(X) in the iteration.

Iteration Func-count min f(x) Procedure

0 1 708.748

1 3 675.4 initial simplex

2 5 645.167 expand

3 7 638.839 expand

4 9 591.374 expand

5 10 591.374 reflect

6 12 554.554 expand

[Vladimir Sovkov]

I do not think it is possible with fminsearch due to a (generally) multidimensional nature of the variable, which is hard to display in a compact form.

The things are easier with the 1D variable: this case you can use an alternative function such as fminbnd, which is able to output the X values at every next itereation via the same approach; besides, it must be faster in the 1D case.