Fmincon
in [Matlab]
|
From: Edgar Galvan on 25 Jun 2010 13:53 Hi all, I am trying to optimize a function function y=myfun(X,C,K) where X is a vector of variables and C and K are vectors of constants so i have [x,fval]=fmincon(@(X)myfun(X,C,K),A,b) for debugging purposes, it would be very helpful if I could tell fmincon to keep certain elements of X constant during optimization. Is there a way to do this easily? Thanks in advance
From: Matt J on 25 Jun 2010 14:23 "Edgar Galvan" <dekuf409(a)aol.com> wrote in message <i02qe0$jr8$1(a)fred.mathworks.com>... > for debugging purposes, it would be very helpful if I could tell fmincon to keep certain elements of X constant during optimization. Is there a way to do this easily? ========= You could do something like the following: Objective=@(X)myfun(X,C,K); xFixed=blablabla; subObjective=@(xVarying) objective([xVarying ; xFixed]); [xVarying,fval]=fmincon(subObjective,[xVaryingInitialGuess;xFixed],A,b) Of course, in your actual situation, the concatenation [xVarying ; xFixed] might have to be substituted with some more complicated interweaving of the two sets of variables. I emphasize that this is only recommendable for debugging purposes! It is NOT efficient to do this in the implementation of an actual algorithm.
From: Edgar Galvan on 25 Jun 2010 15:30 "Matt J " <mattjacREMOVE(a)THISieee.spam> wrote in message <i02s68$g2h$1(a)fred.mathworks.com>... > "Edgar Galvan" <dekuf409(a)aol.com> wrote in message <i02qe0$jr8$1(a)fred.mathworks.com>... > > > for debugging purposes, it would be very helpful if I could tell fmincon to keep certain elements of X constant during optimization. Is there a way to do this easily? > ========= > > > You could do something like the following: > > > Objective=@(X)myfun(X,C,K); > > xFixed=blablabla; > subObjective=@(xVarying) objective([xVarying ; xFixed]); > > [xVarying,fval]=fmincon(subObjective,[xVaryingInitialGuess;xFixed],A,b) > > > Of course, in your actual situation, the concatenation > [xVarying ; xFixed] might have to be substituted with some more complicated interweaving of the two sets of variables. > > I emphasize that this is only recommendable for debugging purposes! It is NOT efficient to do this in the implementation of an actual algorithm. Hum, I can't seem to get it to work. It still varies all the elements in X. This is basically what what my code looks like (simplified). Objective function: function f=myfun(X) f =2*X(1)^2-X(2); Script: A=[1 2;-4 0;0 -1]; b=[5 -7 -2]; Objective=@(X)myfun(X); xFixed=1; subObjective=@(xVarying)Objective([xVarying ; xFixed]); xVaryingInitialGuess=1; [xVarying,fval]=fmincon(subObjective,[xVaryingInitialGuess;xFixed],A,b); Did I do something dumb? Thanks
From: Alan Weiss on 25 Jun 2010 17:01 On 6/25/2010 3:30 PM, Edgar Galvan wrote: > "Matt J " <mattjacREMOVE(a)THISieee.spam> wrote in message > <i02s68$g2h$1(a)fred.mathworks.com>... >> "Edgar Galvan" <dekuf409(a)aol.com> wrote in message >> <i02qe0$jr8$1(a)fred.mathworks.com>... >> >> > for debugging purposes, it would be very helpful if I could tell >> fmincon to keep certain elements of X constant during optimization. Is >> there a way to do this easily? ========= >> >> >> You could do something like the following: >> >> >> Objective=@(X)myfun(X,C,K); >> >> xFixed=blablabla; >> subObjective=@(xVarying) objective([xVarying ; xFixed]); >> >> [xVarying,fval]=fmincon(subObjective,[xVaryingInitialGuess;xFixed],A,b) >> >> >> Of course, in your actual situation, the concatenation >> [xVarying ; xFixed] might have to be substituted with some more >> complicated interweaving of the two sets of variables. >> >> I emphasize that this is only recommendable for debugging purposes! It >> is NOT efficient to do this in the implementation of an actual algorithm. > > Hum, I can't seem to get it to work. It still varies all the elements in > X. This is basically what what my code looks like (simplified). > Objective function: > function f=myfun(X) > f =2*X(1)^2-X(2); > > Script: > A=[1 2;-4 0;0 -1]; b=[5 -7 -2]; > > Objective=@(X)myfun(X); > xFixed=1; > subObjective=@(xVarying)Objective([xVarying ; xFixed]); > xVaryingInitialGuess=1; > > [xVarying,fval]=fmincon(subObjective,[xVaryingInitialGuess;xFixed],A,b); > > Did I do something dumb? > > Thanks Another approach: Use the interior-point algorithm. Set equal upper and lower bounds for the components you want fixed. It's inefficient, but very easy. Alan Weiss MATLAB mathematical toolbox documentation
From: Matt J on 25 Jun 2010 17:39 "Edgar Galvan" <dekuf409(a)aol.com> wrote in message <i0304n$64b$1(a)fred.mathworks.com>... > Objective function: > function f=myfun(X) > f =2*X(1)^2-X(2); > > Script: > A=[1 2;-4 0;0 -1]; b=[5 -7 -2]; > > Objective=@(X)myfun(X); > xFixed=1; > subObjective=@(xVarying)Objective([xVarying ; xFixed]); > xVaryingInitialGuess=1; > > [xVarying,fval]=fmincon(subObjective,[xVaryingInitialGuess;xFixed],A,b); > > Did I do something dumb? ========= No, my mistake. I think Alan's approach is probably the easiest, but here's the necessary modification: N=length(xVaryingInitialGuess); Asub=A(:,1:N); bsub=b-A(:,N+1:end)*xFixed; [xVarying,fval]=fmincon(subObjective,xVaryingInitialGuess,Asub,bsub);
|
Pages: 1 Prev: textscan doesn't scan data, but does scan similar data Next: How do you understand accumarray? |