fminsearch taking too much time - results do not show up yet
5 views (last 30 days)
Show older comments
%
function A=obj(b,Y, YL, tempcyc, cyc, Yzero, age, agesq, educ, ageav, agesqav, educav, qv, year1, year2, year3, year4, year5, year6, year7, workregion1, workregion2, workregion3, workregion4, workregion5, workregion6, workregion7, workregion8, workregion9, workregion10, workregion11, workregion12, workregion13, workregion14, workregion15, workregion16, qw)
%
A=0;
for i=1:715
S=0;
for m=1:12
P=1;
for t=1:8
P=P*(normcdf((2*Y(i,t)-1)*(b(1)*YL(i,t)+b(2)*tempcyc(i,t)+b(3)*cyc(i,t)+b(4)*Yzero(i,t)+b(5)*age(i,t)+b(6)*agesq(i,t)+b(7)*educ(i,t)+b(8)*ageav(i,t)+b(9)*agesqav(i,t)+b(10)*educav(i,t)+b(11)*1+b(12)*sqrt(2)*qv(m,1)+b(13)*year1(i,t)+b(14)*year2(i,t)+b(15)*year3(i,t)+b(16)*year4(i,t)+b(17)*year5(i,t)+b(18)*year6(i,t)+b(19)*year7(i,t)+b(20)*workregion1(i,t)+b(21)*workregion2(i,t)+b(22)*workregion3(i,t)+b(23)*workregion4(i,t)+b(24)*workregion5(i,t)+b(25)*workregion6(i,t)+b(26)*workregion7(i,t)+b(27)*workregion8(i,t)+b(28)*workregion9(i,t)+b(29)*workregion10(i,t)+b(30)*workregion11(i,t)+b(31)*workregion12(i,t)+b(32)*workregion13(i,t)+b(33)*workregion14(i,t)+b(34)*workregion15(i,t)+b(35)*workregion16(i,t))));
end
%
S=S+qw(m,1)*P;
end
%
A=A+log(S/sqrt(pi));
A=-A;
end
This is my objective function I am minimizing (actually maximizing; I am minimizing -A) and the parameters I am estimating is b=[b(1)............b(35)]. Y, YL, tempcyc,.........., qw are matrices that I am imported as data. The objective function consists of a product(across t=1:8T) nested within a sum (across m=1:12) that is in turn nested within a sum (across n=1:715). And below is my code using fminsearch for my unconstrained minimization.
% %%Minimization
start=[0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0];
iter=50000;
options=optimset('Display','iter','MaxIter',iter,'MaxFunEvals',100000);
[b, fval, exitflag, output]=fminsearch(@(b)obj(b,Y, YL, tempcyc, cyc, Yzero, age, agesq, educ, ageav, agesqav, educav, qv, year1, year2, year3, year4, year5, year6, year7, workregion1, workregion2, workregion3, workregion4, workregion5, workregion6, workregion7, workregion8, workregion9, workregion10, workregion11, workregion12, workregion13, workregion14, workregion15, workregion16, qw), start, options);%
%%Results
fprintf('[b] : % 1.4e % 1.4e %1.4e % 1.4e % 1.4e % 1.4e \n',...
b(1),b(2),b(3),b(4), b(5),b(6),b(7),b(8), b(9),b(10),b(11),b(12), b(13),b(14),b(15),b(16), b(17),b(18),b(19),b(20), b(21),b(22),b(23),b(24), b(25),b(26),b(27),b(28), b(29),b(30),b(31),b(32), b(33),b(34), b(35));
end
The problem is that the optmization results do not show up even after 12+ hours (It is still reflecting, expanding, contracting inside, etc.) Could someone give me an idea on how to make the process faster?
Thank you.
0 Comments
Answers (1)
Matt J
on 20 Jun 2014
Edited: Matt J
on 20 Jun 2014
35 unknown variables is way too big for fminsearch. It's theoretically guaranteed to converge for only 1 unknown, and gives good empirical results for up to about 6.
You will probably need an Optimization Toolbox solver.
1 Comment
Matt J
on 20 Jun 2014
Edited: Matt J
on 20 Jun 2014
The objective function consists of a product(across t=1:8T) nested within a sum (across m=1:12) that is in turn nested within a sum (across n=1:715). And below is my code using fminsearch for my unconstrained minimization.
Such operations are easily vectorized using MATLAB commands. You should consider doing so to improve speed-efficiency of your code. It looks like the whole thing boils down to something of the form
G=reshape( prod(normcdf(M*b) ,2) , 12,715);
S=sum(qw.'*G);
for an appropriate matrix M.
See Also
Categories
Find more on Dimensionality Reduction and Feature Extraction in Help Center and File Exchange
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!