Different inputs for x0 at fmincon

I am using a fmincon function in Matlab because I want to optmize a function with constraints.

I wrote: [x,fval] = fmincon(@time,x0,[],[],[],[],[],[],@confun);

where [t,range,W_total]=time(x0)

The problem is that the inputs that I want to send to the constraints function (confun) are not the same that I put at X0 in order to solve the function time. I would like that the parameters that I obtained as outputs of the function time (t,range,W_total) were the inputs for the constrain function.

How can I solve this problem?

Answer 1

[Walter Roberson]

Write a little wrapper function on the constraint function that takes an x as input, calls your "time" on that x, gets the outputs, and proceeds with those values as inputs to your real constraint function.

Answer 2

[Matt J]

Your objective function time() and your constraint function confun() have to be written as separate functions of x, and there's no way to make fmincon derive output for both of them from a single call to a 3rd function. You could make wrappers as Walter suggests, but this still means that multiple calls to time() would be made by fmincon for the same x0, and you will have to accept that.

That said, however, there are some fancy tricks with which you can make the objective and constraint functions share certain intermediate quantities that you don't want to recompute redundantly,

http://www.mathworks.com/matlabcentral/newsreader/view_thread/269936#877852

Since this is a low-dimensional problem, however, and since you are a novice user (judging from your earlier posts), I don't know if it's worth starting at this level of complexity.

[Romans]

I have done it the way you told me but nothing happens. I mean in every iteration x0 is the same so the values that get into the constrain function are the same: I wrote:

x0=[5,60,85,0.3];
[x,fval] = fmincon(@time,x0,[],[],[],[],[],[],@confun);

and every time obviously inside confun there is the same vector x0. How do I have to do it?

[Romans]

I just want an alternative of using:

for i1=1 to 10 
 for i2=2 to 10 
  for i3=1 to 10
    [CALCULATIONS]
  end
 end 
end

[Matt J]

I don't understand what you mean when you say "in every iteration x0 is the same so the values that get into the constrain function are the same". What evidence do you see of this?

And did you rewrite confun as Walter suggested? Does it take x as input or something else?

[Sean de Wolski]

What range of values is your objective function returning and what sensitivity is it? If your objective function is only sensitive to large changes in x, use the 'DiffMinChange' Option in optimset to force fmincon() to use a larger step in its finite differences.

This of course comes after you have fixed everything Walter and Matt have suggested.

[Walter Roberson]

If x0 is the same for every iteration, what is different about the iterations?

[Romans]

Yes I did the changes that he suggested, I calculate again the 3 functions that I need as constraints on the constrain function. I have done the examples that there are on the mathwork page, and x0 is different in every 3-4 iterations. But in my case, every time x0 is the same. I will use the optimset that Sean suggested in order to see if it works, but the result should be very close to the one that I initially stated, this is why is weird, because I know that the result should be something like [5,60,85,0.3] and this is the value that x0 has every time, so I do not think that i am so lucky to have inserted the best value initially.

[Matt J]

Your terminology is confusing. The notation x0 usually refers to the initial guess of the unknown vector that you feed fmincon, not the solution it returns.

We also need a lot more information to know what's going on:

What initial starting guess x0 did you provide? How many iterations did it run? What was the final x value returned? What other diagnostic information did fmincon print to the screen? When you run with 3 output arguments

   [x,fval,exitflag]=fmincon(...)

what is the value of exitflag that gets returned?

[Sean de Wolski]

Get all four!

    [X,FVAL,EXITFLAG,OUTPUT] = fmincon(FUN,X0,...)

[Romans]

%MAIN FUNCTION
x0=[5,60,85,0.3];
[x,fval,exitflag,t] = fmincon(@time,x0,[],[],[],[],[],[],@confun);
---------------------------
function [t]=time(val)
A=val(1);
S=val(2);
P_i=val(3);
Cl=val(4);
[CALCULATIONS THAT BRINGS OUT variable t with A,S,P_i,Cl]
----------------------------

function [c,ceq]=confun(val2)
[CALCULATIONS to find variables: roc,range,W_total and to_dist]
%Nonlinear inequality constraints
c=[500-roc; 2200-range; W_total-1102; to_dist-1000];
%Nonlinear equality constraints
ceq=[];

[Romans]

This is the actual body of my program. The idea is to calculate with the four parameters that I insert as inputs in x0 (prediction) all kind of calculations and finally find a variable called t, which is the one that I want to optimize. I want the t min with the constraints in the confun function.

I do really appreciate your efforts to help me because I am doing my Master of Science thesis and I had never done anything with optimization functions like this one.

[Matt J]

The code you've shown doesn't answer the questions we asked you. I will repeat them:

What initial starting guess x0 did you provide? How many iterations did it run? What was the final x value returned? What other diagnostic information did fmincon print to the screen? When you run with 3 output arguments

   [x,fval,exitflag,output]=fmincon(...)

what is the value of exitflag and output that gets returned?

Also, you haven't really shown the part of your code that is informative. Things you've hidden like [CALCULATIONS THAT BRINGS OUT variable t with A,S,P_i,Cl] are the part of the code that carries the real information.

[Romans]

I cannot put all my code here because it is too long, but if you give me your mail I would really appreciate it