Prev: compile lap.cpp
Next: blstheta(Price,Strike,Int_Rate,Time,Volatility)
From: Luca Zanotti Fragonara on 28 May 2010 05:30
Hello, I'm trying to solve a minimisation problem with fmincon. The only trouble is that I want that one of the parameter of my function should assume only integer value.

My code is something like that, where par0S is a vector of parameters. I want the ONE of the value of par0S assumes only integer values.

fitness=@(par0S)err_tf_ss_NL(par0S,parmin,parmax,M,disp_meas,vel_meas,acc_meas,inp_meas,f_meas,istante,dt,H,part_path);
nonlcon=@(par)mycon(par,max(f_meas),parmin,parmax);
[par(:,1), Fob(1,1), flag(1,1)]=patternsearch(fitness,par0S,[],[],[],[],parminS,parmaxS,nonlcon,opt);


I thought that in my fitness function err_tf_ss_NL, i could put a "round" on par(ii), if ii is the index that identifies my "integer" parameter. But that seems to not work properly. I think that in this case the fmincon calculates a bad Jacobian matrix, resulting the derivative of fitness respect par(ii) "straigth", but actually it's not: he must calculates the derivative with the next integer value.

Do you have any suggestion?

Thank you in advance,

Luca
From: John D'Errico on 28 May 2010 05:41
"Luca Zanotti Fragonara" <Luca_Zanotti(a)libero.it> wrote in message <hto2fe$drr$1(a)fred.mathworks.com>...
> Hello, I'm trying to solve a minimisation problem with fmincon. The only trouble is that I want that one of the parameter of my function should assume only integer value.
>
> My code is something like that, where par0S is a vector of parameters. I want the ONE of the value of par0S assumes only integer values.
>
> fitness=@(par0S)err_tf_ss_NL(par0S,parmin,parmax,M,disp_meas,vel_meas,acc_meas,inp_meas,f_meas,istante,dt,H,part_path);
> nonlcon=@(par)mycon(par,max(f_meas),parmin,parmax);
> [par(:,1), Fob(1,1), flag(1,1)]=patternsearch(fitness,par0S,[],[],[],[],parminS,parmaxS,nonlcon,opt);
>
>
> I thought that in my fitness function err_tf_ss_NL, i could put a "round" on par(ii), if ii is the index that identifies my "integer" parameter. But that seems to not work properly. I think that in this case the fmincon calculates a bad Jacobian matrix, resulting the derivative of fitness respect par(ii) "straigth", but actually it's not: he must calculates the derivative with the next integer value.
>
> Do you have any suggestion?

NO. You cannot round your parameter inside the
objective function. Fmincon assumes that your
function is a differentiable one. Does this create a
differentiable function? NO!!!

You cannot use fmincon to solve this problem.
Look on the file exchange for an alternative.

John
From: Luca Zanotti Fragonara on 28 May 2010 06:42
"John D'Errico" <woodchips(a)rochester.rr.com> wrote in message <hto33f$m8h$1(a)fred.mathworks.com>...
> "Luca Zanotti Fragonara" <Luca_Zanotti(a)libero.it> wrote in message <hto2fe$drr$1(a)fred.mathworks.com>...
> > Hello, I'm trying to solve a minimisation problem with fmincon. The only trouble is that I want that one of the parameter of my function should assume only integer value.
> >
> > My code is something like that, where par0S is a vector of parameters. I want the ONE of the value of par0S assumes only integer values.
> >
> > fitness=@(par0S)err_tf_ss_NL(par0S,parmin,parmax,M,disp_meas,vel_meas,acc_meas,inp_meas,f_meas,istante,dt,H,part_path);
> > nonlcon=@(par)mycon(par,max(f_meas),parmin,parmax);
> > [par(:,1), Fob(1,1), flag(1,1)]=patternsearch(fitness,par0S,[],[],[],[],parminS,parmaxS,nonlcon,opt);
> >
> >
> > I thought that in my fitness function err_tf_ss_NL, i could put a "round" on par(ii), if ii is the index that identifies my "integer" parameter. But that seems to not work properly. I think that in this case the fmincon calculates a bad Jacobian matrix, resulting the derivative of fitness respect par(ii) "straigth", but actually it's not: he must calculates the derivative with the next integer value.
> >
> > Do you have any suggestion?
>
> NO. You cannot round your parameter inside the
> objective function. Fmincon assumes that your
> function is a differentiable one. Does this create a
> differentiable function? NO!!!
>
> You cannot use fmincon to solve this problem.
> Look on the file exchange for an alternative.
>
> John

Okay, thank you, I was pretty sure that there was no alternative, by the way, thank you again.

Luca
 | 
Pages: 1
Prev: compile lap.cpp
Next: blstheta(Price,Strike,Int_Rate,Time,Volatility)