How to do sort rows more efficiently.

Hi, all!

I plan to sort rows to an numerical matrix. First sort them in descending order based on first column, then if there is a tie, sort those in the tie in descending order on second column, and so on(if tie sort on the next column).

Generally, it is possible to do this kinder of sorting but I notice "sortrows" command and it can sort rows based on which column. But is there a easier way to do my algorithm described above?

Thanks.

Answer 1

[Jan]

I cannot imagine that there is any easier method than sortrows:

x = randi(4, 8, 8);
y = sortrows(x);

For more performance see: http://www.mathworks.com/matlabcentral/answers/57629

[EDITED] For large arrays like randi(4, 1e5, 10) this is about 10% faster than sortrows:

function [y, index] = leanSortRows(x)
[v, index] = sort(x(:, n));          %#ok<ASGLU>
for k = n-1:-1:1
   [v, index2] = sort(x(index, k));  %#ok<ASGLU>
   index       = index(index2);
end
y = x(index, :);

This is similar to the fallback for backward sorting. The MEX-function sortrows.c, which called for standard cases uses the quicksort algorithm of the C-libs, which are obviously slower than Matlab's built-in SORT.

[C Zeng]

Thanks, Jan! Let me think it over again.