Mercurial > hg > camir-aes2014
annotate toolboxes/distance_learning/mlr/loss/lossHingeFullMKL.m @ 0:e9a9cd732c1e tip
first hg version after svn
| author | wolffd |
|---|---|
| date | Tue, 10 Feb 2015 15:05:51 +0000 |
| parents | |
| children |
| rev | line source |
|---|---|
| wolffd@0 | 1 function Xi = lossHingeMKLFull(W, Psi, M, gradient) |
| wolffd@0 | 2 % |
| wolffd@0 | 3 % Xi = lossHingeMKLFull(W, Psi, M, gradient) |
| wolffd@0 | 4 % |
| wolffd@0 | 5 % W: d*d*m metric |
| wolffd@0 | 6 % Psi: d*d*m feature matrix |
| wolffd@0 | 7 % M: the desired margin |
| wolffd@0 | 8 % gradient: if 0, returns the loss value |
| wolffd@0 | 9 % if 1, returns the gradient of the loss WRT W |
| wolffd@0 | 10 |
| wolffd@0 | 11 m = size(W, 3); |
| wolffd@0 | 12 |
| wolffd@0 | 13 Xi = M; |
| wolffd@0 | 14 for i = 1:m |
| wolffd@0 | 15 Xi = Xi - sum(sum(W(:,:,i) .* Psi(:,:,i))); |
| wolffd@0 | 16 end |
| wolffd@0 | 17 Xi = max(0, Xi); |
| wolffd@0 | 18 |
| wolffd@0 | 19 if gradient & Xi > 0 |
| wolffd@0 | 20 Xi = -Psi; |
| wolffd@0 | 21 end |
| wolffd@0 | 22 end |