Memory efficient vectorization of a for loop

2 views (last 30 days)
Haider Ali
Haider Ali on 29 Mar 2020
Answered: darova on 29 Mar 2020
Hello all,
I have the following piece of code which generate a nxnxm logical matrix A at the end. A matrix contains a lot of zeros. Each nxn matrix contains n 1s and remaining 0s.
clc;
clear; close all;
n = 3;
M = 2*n;
No_of_switches_in_each_column = [2,3,3,2,2];
tic
Ns = sum(No_of_switches_in_each_column);
All_possible_switching_states = 2^Ns;
I = eye(6);
Qi = [1,3,5,2,4,6];
Map_s2c = I(Qi',:);
Map_c2s = I(:,Qi);
TT_4allpossible_switch_states = dec2bin(0:All_possible_switching_states-1); % Truth table for Ns switches
Sw = cell(1,Ns); % Cell to store state (bar or cross) matricess of Ns No of switches.
A = false([6 6 All_possible_switching_states]);
%//********************************************************************//
for i = 1:All_possible_switching_states
comb_i = TT_4allpossible_switch_states(i,:);
for n = 1:Ns
if(strcmp(comb_i(n),'0'))
Sw{n} = [1 0; 0 1];
else
Sw{n} = [0 1; 1 0];
end
end
S1 = Sw{1}; S2 = Sw{2}; S3 = Sw{3};
S4 = Sw{4}; S5 = Sw{5}; S6 = Sw{6};
S7 = Sw{7}; S8 = Sw{8}; S9 = Sw{9};
S10 = Sw{10}; S11 = Sw{11};
S12 = Sw{12};
C1 = logical([1 zeros(1,5);...
zeros(2,1) S1 zeros(2,3);...
zeros(2,3) S2 zeros(2,1);...
zeros(1,5) 1]);
C2 = logical([S3 zeros(2,4);...
zeros(2) S4 zeros(2);...
zeros(2,4) S5]);
C3 = logical([S6 zeros(2,4);...
zeros(2) S7 zeros(2);...
zeros(2,4) S8]);
C4 = logical([S9 zeros(2,4);...
zeros(1,2) 1 zeros(1,3);...
zeros(1,3) 1 zeros(1,2);...
zeros(2,4) S10]);
C5 = logical([1 zeros(1,5);...
zeros(2,1) S11 zeros(2,3);...
zeros(2,3) S12 zeros(2,1);...
zeros(1,5) 1]);
A(:,:,i) = logical(C1*C2*Map_s2c*C3*Map_c2s*C4*C5);
end
I want an efficient implementation of the above code in terms of memory (it fails for larger arrays - array exceeds maximum array size preference) and computations.
It would be desirable to not use for loop and cell arrays.
Thanks
  2 Comments
Haider Ali
Haider Ali on 29 Mar 2020
@darova, yes I have looked at sparse matrix. But only 2D sparse matrices are supported.
I was thinking of first removing the for loop and then go for sparse matrix.

Sign in to comment.

Answers (1)

darova
darova on 29 Mar 2020
Store values manually like sparse
% preallocation?
irow = [];
icol = [];
iplane = [];
for i = 1:...
% do stuff
val = logical(C1*C2*Map_s2c*C3*Map_c2s*C4*C5);
if sum(val(:))
[ii,jj] = find(val);
irow = [irow; ii];
icol = [icol; jj];
iplane = [iplane; i+ii*0];
end
end

Products


Release

R2019b

Community Treasure Hunt

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

Start Hunting!