MATLAB Answers

Multidimensional operations without for loop

2 views (last 30 days)
Aram Eskandari
Aram Eskandari on 27 Mar 2020
Commented: Aram Eskandari on 27 Mar 2020
I have a vector and an input
y = 8.3; % this is different for each run, but not relevant for this question
maxloc = [1 2 1 3 3]; % n long, in this case n = 5, each value goes from 1 to d, in this case d = 3
And a matrix which is attached in .mat-format, it is a double matrix with shape given below
size(theta) = (3,8,5); % d=3, n=5, the columns however are always 8
Which I operate on in the following way
for i=1:n % n = 5 in this case
theta(maxloc(i),7,i) = theta(maxloc(i),7,i) + 0.5;
theta(maxloc(i),8,i) = theta(maxloc(i),8,i) + ((y-theta(maxloc(i),3,i)).^2)*0.5;
theta(maxloc(i),6,i) = 1./(gamrnd(theta(maxloc(i),7,i),1./theta(maxloc(i),8,i)));
theta(maxloc(i),5,i)= theta(maxloc(i),5,i) + 1./theta(maxloc(i),6,i);
theta(maxloc(i),4,i)= (theta(maxloc(i),5,i).*theta(maxloc(i),4,i)+(y./theta(maxloc(i),6,i)) )./(1./theta(maxloc(i),6,i)+theta(maxloc(i),5,i));
theta(maxloc(i),3,i) = normrnd(theta(maxloc(i),4,i),sqrt(1./(theta(maxloc(i),5,i)+1./theta(maxloc(i),6,i))));
end
As you can imagine, the code gets really slow as n increases, so I want to get rid of the for loop.
I haven't included columns 1,2 here because I have a solution for those columns without the use of for loops. However I'm not quite sure how I can get these operations to work in a way which excludes the loops, any help would be greatly appreciated.
EDIT:
The reason this takes so long is because I have 50.000 data-points, meaning y is really 50.000 long and I want to run through my algorithm for each y, and preferably for a large n.

  0 Comments

Sign in to comment.

Accepted Answer

Matt J
Matt J on 27 Mar 2020
Edited: Matt J on 27 Mar 2020
Q=nan(8,n);
[J,K]=ndgrid(1:8,1:n);
I=maxloc(K);
thetaIndices=sub2ind(size(theta), I,J,K);
QIndices=sub2ind(size(Q), J,K);
Q(QIndices)=theta(thetaIndices);
Q(7,:) = Q(7,:) + 0.5;
Q(8,:) = Q(8,:) + ((y-Q(3,:)).^2)*0.5;
Q(6,:) = 1./(gamrnd(Q(7,:),1./Q(8,:)));
tmp=1./Q(6,:);
Q(5,:)= Q(5,:) + tmp;
tmp2=tmp+Q(5,:);
Q(4,:)= (Q(5,:).*Q(4,:)+(y./Q(6,:)) )./(tmp2);
Q(3,:) = normrnd(Q(4,:),sqrt(1./(tmp2)));
theta(thetaIndices)=Q(QIndices);

  4 Comments

Show 1 older comment
Aram Eskandari
Aram Eskandari on 27 Mar 2020
I have edited my question to include the reason why it takes so long to run.
Matt J
Matt J on 27 Mar 2020
See my edited version, which is loop-less.
Aram Eskandari
Aram Eskandari on 27 Mar 2020
Thanks so much! This seems to be a lot quicker without the for loops, especially with a lot of datapoints. Giving you the green tick.

Sign in to comment.

More Answers (0)

Products


Release

R2019b

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!