How to speed up the "unique" function when sorting out unique columns

7 次查看（过去 30 天）

Gunnar 2015-3-18

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/183821-how-to-speed-up-the-unique-function-when-sorting-out-unique-columns

评论： FirefoxMetzger 2016-8-15

As you can see by running the example code below, the time it takes to compute the number of unique columns quickly deteriorates with matrix size.

Others have proposed ways to write inline code in order to speed this up, but those solutions do not seem to cover column-wise (alt. row-wise) uniqueness.

Please help.

clear all, close all
n = 20; % #objects
% compute all possible combinations of 0 and 1 (#combinations = 2^n)
% 0 - object is turned off
% 1 - object is active
% furthermore, as an example the states 01 and 10 are assumed to be equivalent since order is not an issue
combs = [0 1];
for i = 2:n
    combs = combvec(combs,[0 1]);
    combsSort = sort(combs, 1); % sort columnwise
  tic
    uniqueCombs = unique(combsSort.','rows').'; % find unique columns by transposing twice
    toc
    size(uniqueCombs)
end

1 个评论
显示 -1更早的评论隐藏 -1更早的评论

FirefoxMetzger 2016-8-15

在 MATLAB Online 中打开

This entire sorting can be solved by:

result = ones(N,N);
result = fliplr(tril(result));

To suggest a better algorithm (if still needed) please be more specific about what you want. The unique() command is the most efficient way to solve a general problem. However, there is often a more efficient implementation depending on the need.

请先登录，再进行评论。

请先登录，再回答此问题。