下面的循环可以进行矢量化吗?循环中的每次迭代形成一个外部产品,然后对称并将结果存储为矩阵中的列。预计m
很大(例如,1e4),而s
很小(例如,10)。
% U and V are m-by-s matrices
A = zeros(s^2, m); % preallocate
for k = 1:m
Ak = U(k,:)' * V(k,:);
Ak = (Ak + Ak')/2;
A(:, k) = Ak(:);
end
修改
以下是3种不同方法的比较:迭代大维m
,迭代小维s
和基于bsxfun
的解决方案(接受且最快的答案) )。
s = 5; m = 100000;
U = rand(m, s);
V = rand(m, s);
% Iterate over large dimension
tic
B = zeros(s^2, m);
for k = 1:m
Ak = U(k,:)' * V(k,:);
Ak = (Ak + Ak')/2;
B(:, k) = Ak(:);
end
toc
% Iterate over small dimension
tic
A = zeros(s, s, m);
for i = 1:s
A(i,i,:) = U(:, i) .* V(:, i);
for j = i+1:s
A(i,j,:) = (U(:,i).*V(:,j) + U(:, j).*V(:, i))/2;
A(j,i,:) = A(i,j,:);
end
end
A = reshape(A, [s^2, m]);
toc
% bsxfun-based solution
tic
A = bsxfun( @times, permute( U, [1 3 2] ), permute( V, [ 1 2 3 ] ) );
A = .5 * ( A + permute( A, [1 3 2] ) );
B = reshape( A, [m, s^2] )';
toc
以下是时间比较:
Elapsed time is 0.547053 seconds.
Elapsed time is 0.042639 seconds.
Elapsed time is 0.039296 seconds.
答案 0 :(得分:1)
使用bsxfun
(这是如何完成的,有很多乐趣):
% the outer product
A = bsxfun( @times, permute( U, [1 3 2] ), permute( V, [ 1 2 3 ] ) );
% symmetrization
A = .5 * ( A + permute( A, [1 3 2] ) );
% to vector (per product)
B = reshape( A, [m s^2] )';
基准测试结果(我的机器):
原创方法(迭代大昏暗):
Elapsed time is 0.217695 seconds.
“新”方法(迭代较小的暗淡):
Elapsed time is 0.037538 seconds.
有趣bsxfun
:
Elapsed time is 0.021507 seconds.
正如您所看到的,bsxfun
需要约2/3 - 最快循环的1/2时间......
bsxfun不是很有趣吗?