Repmat is not a built-in function in Matlab. It is slow. An alternative is to do “repmat” by matlab indexing, which is faster.
x=[1 2 3 4]' ; x(:, ones(4,1));%repmat(x, [1 4]) ; x=[1 2 3 4]; x(ones(1,4),:);%repmat(x,[4 1]) |
A more detailed tutorial is found here.
I don’t really agree it’s faster than repmat. Using Steve’s timeit function (http://www.mathworks.com/matlabcentral/fileexchange/18798):
>> x=(1:1000)’;
>> a=@()repmat(x,[1,4000]);
>> b=@()x(:,ones(4000,1));
>> isequal(a(),b())
ans =
1
>> timeit(a)
ans =
0.0105
>> timeit(b)
ans =
0.0105
If you don’t want to download timeit, you can see the same effect by making the operation larger and using tic and toc:
>> x=(1:10000)’;
>> tic;repmat(x,[1,40000]);toc
Elapsed time is 2.493900 seconds.
>> tic;x(:,ones(40000,1));toc
Elapsed time is 2.489806 seconds.
repmat actually does that same indexing. There’s a little overhead in reading in the function, which happens only the first time you use it, and a little bit of parsing of the input, which is a very, very small overhead in any of the cases where you’d worry about speed. For very small arrays you can see a difference:
>> a=@()repmat(x,[1,4]);
>> b=@()x(:,ones(4,1));
>> timeit(a)
ans =
3.6788e-05
>> timeit(b)
ans =
1.4794e-05
repmat makes the code easier to follow. Is 20us really worth using more obscure code?
Further discussion on bsxfun and repmat:
http://blogs.mathworks.com/loren/2008/08/04/comparing-repmat-and-bsxfun-performance/
You must be logged in to post a comment.
