comparison toolboxes/distance_learning/mlr/loss/lossHingeFullMKL.m @ 0:e9a9cd732c1e tip

first hg version after svn

+function Xi = lossHingeMKLFull(W, Psi, M, gradient)
+%
+%   Xi = lossHingeMKLFull(W, Psi, M, gradient)
+%   
+%   W:          d*d*m metric
+%   Psi:        d*d*m feature matrix
+%   M:          the desired margin
+%   gradient:   if 0, returns the loss value
+%               if 1, returns the gradient of the loss WRT W
+    
+    m = size(W, 3);
+
+    Xi = M;
+    for i = 1:m
+        Xi = Xi -  sum(sum(W(:,:,i) .* Psi(:,:,i)));
+    end
+    Xi = max(0, Xi);
+
+    if gradient & Xi > 0
+        Xi = -Psi;
+    end
+end