[Octave-patch-tracker] [patch #8959] pkg-statistics: Replace kmeans with more efficient implementation

[Michael Bentley]

Follow-up Comment #3, patch #8959 (project octave):

Also of note, I have some timing data if people want to see *how much* this is
improved.

----------------------------------------------------------------
| k | Mine in Octave | Octave's   | Mine in Matlab | Matlab's  |
|---|----------------|------------|----------------|-----------|
| 3 | 0.197 sec      | 12.268 sec | 0.313 sec      | 0.245 sec |
| 4 | 0.578 sec      | 17.961 sec | 0.629 sec      | 0.475 sec |
| 5 | 0.458 sec      | 25.572 sec | 0.614 sec      | 0.402 sec |
| 6 | 1.026 sec      | 50.141 sec | 1.160 sec      | 0.739 sec |
| 7 | 1.237 sec      | 41.985 sec | 1.418 sec      | 0.534 sec |
| 8 | 1.077 sec      | 21.745 sec | 1.289 sec      | 0.929 sec |
----------------------------------------------------------------

This timing experiment was done on a grid of 128x128 elements, each one having
16 channels.  I call it with

    [m,n,p] = size(params);
    [elements, clusters] = kmeans(reshape(params, [m*n, p]), k);
    elements = reshape(elements, [m,n]);



    _______________________________________________________

Reply to this item at:

  <http://savannah.gnu.org/patch/?8959>

_______________________________________________
  Message sent via/by Savannah
  http://savannah.gnu.org/