Mercurial > hg > machine-learning-hw7
changeset 6:6d94b2bafcd1 default tip
Replace complicated memory-intensive operation with a faster loop
| author | Jordi Gutiérrez Hermoso <jordigh@octave.org> |
|---|---|
| date | Tue, 06 Dec 2011 11:49:32 -0500 |
| parents | 87433ad16bbf |
| children | |
| files | findClosestCentroids.m |
| diffstat | 1 files changed, 11 insertions(+), 3 deletions(-) [+] |
line wrap: on
line diff
--- a/findClosestCentroids.m +++ b/findClosestCentroids.m @@ -5,9 +5,17 @@ ## vector of centroid assignments (i.e. each entry in range [1..K]) ## - ## Using broadcasting (auto BSX) as available in Octave 3.5.0+ - d = sum ((permute (X, [1,3,2]) - permute (centroids, [3,1,2])).^2, 3); + d = zeros (rows (X), rows (centroids)); + + ## This loop can be avoided with greater memory usage, but it doesn't + ## seem to speed things up. + for i = 1:rows (centroids) + + ## Using broadcasting (auto BSX) as available in Octave 3.5.0+ + d(:, i) = sum ((X - centroids(i,:)).^2, 2); + + endfor + [~, idx] = min (d, [], 2); endfunction -