フィルターのクリア

How to speed up calculation taking inputs from 2 separate and very large matrices (working code below)

1 回表示 (過去 30 日間)
klb
klb 2018 年 10 月 25 日
編集済み: per isakson 2018 年 10 月 28 日
Hi. The code below works. updated portion the REAL is 'a' is at 10600X5 and and 'b' at 60X5 and takes forever to run. How can I massively speed up the calculation?
*updated*
tic
a = randi([0,100],10600,5) % example only to run the code
b = randi([0,20],60,5)
new = zeros(size(a,1)*size(b,1),size(a,2));
w = 0;
for s= 1:size(a,1)
for t= 1:size(b,1)
new(w+t,:)= [(3*a(s,1)+b(t,1)), (5*a(s,2)+b(t,2)), (1*a(s,3)+b(t,3)), (5*a(s,4)+b(t,4)) , (2*a(s,5)+b(t,5))];
end
w=w+size(b,1);
end
toc
  4 件のコメント
Jan
Jan 2018 年 10 月 26 日
@klb: Do I understand correctly: The posted code works with the needed speed, but some other code, which you do not show, is much slower? We cannot know the reason of this difference without seeing the other code. So all I can do is to suggest letting the profile find out, where the time is spent.
per isakson
per isakson 2018 年 10 月 28 日
"massively speed up" How much is that?

サインインしてコメントする。

回答 (1 件)

per isakson
per isakson 2018 年 10 月 28 日
編集済み: per isakson 2018 年 10 月 28 日
Try
>> cssm()
Elapsed time is 0.073258 seconds.
Elapsed time is 0.073459 seconds.
Elapsed time is 0.042250 seconds.
Elapsed time is 0.019368 seconds.
ans =
1×3 logical array
1 1 1
where
function cssm()
a = randi([0,100],10800,5); % example only to run the code
b = randi([0,20],60,5);
tic, new1 = cssm_1( a, b ); toc
tic, new2 = cssm_2( a, b ); toc
tic, new3 = cssm_3( a, b ); toc
tic, new4 = cssm_4( a, b ); toc
[ all(new1(:)==new2(:)), all(new1(:)==new3(:)), ll(new1(:)==new4(:)) ]
end
function new = cssm_1( a, b )
%%*updated*
new = zeros(size(a,1)*size(b,1),size(a,2));
w = 0;
for s = 1:size(a,1)
for t = 1:size(b,1)
new(w+t,:)= [(3*a(s,1)+b(t,1)) ...
,(5*a(s,2)+b(t,2)) ...
,(1*a(s,3)+b(t,3)) ...
,(5*a(s,4)+b(t,4)) ...
,(2*a(s,5)+b(t,5)) ];
end
w=w+size(b,1);
end
end
function new = cssm_2( a, b )
new = zeros(size(a,1)*size(b,1),size(a,2));
w = 0;
a = [3,5,1,5,2].*a;
for s = 1:size(a,1)
for t = 1:size(b,1)
new(w+t,:)= [(a(s,1)+b(t,1)) ...
,(a(s,2)+b(t,2)) ...
,(a(s,3)+b(t,3)) ...
,(a(s,4)+b(t,4)) ...
,(a(s,5)+b(t,5)) ];
end
w=w+size(b,1);
end
end
function new = cssm_3( a, b )
new = zeros(size(a,1)*size(b,1),size(a,2));
w = 0;
a = [3,5,1,5,2].*a;
for s = 1:size(a,1)
for t = 1:size(b,1)
new(w+t,:)= a(s,:)+b(t,:);
end
w=w+size(b,1);
end
end
function new = cssm_4( a, b )
new = zeros(size(a,1)*size(b,1),size(a,2));
w = size(b,1);
a = [3,5,1,5,2].*a;
for s = 1:size(a,1)
new( ( 1+(s-1)*w : s*w ), : ) = a(s,:)+b;
end
end
Caveat: I increased 10600 to 10800 to make it a multiple of 60.

カテゴリ

Help Center および File ExchangeClocks and Timers についてさらに検索

Community Treasure Hunt

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

Start Hunting!

Translated by