Added nnpredict to make predicting easier.

Author: rasmusbergpalm

NN/nnpredict.m

@@ -0,0 +1,12 @@
+function labels = nnpredict(nn, x)
+    if nn.normalize_input==1;
+       x = (x-repmat(nn.normalizeMean,size(x,1),1))./repmat(nn.normalizeStd,size(x,1),1);
+    end
+    
+    nn.testing = 1;
+    nn = nnff(nn, x, zeros(size(x,1), nn.size(end)));
+    nn.testing = 0;
+    
+    [~, i] = max(nn.a{end},[],2);
+    labels = i;
+end

NN/nntest.m

@@ -1,14 +1,6 @@
 function [er, bad] = nntest(nn, x, y)
-    if nn.normalize_input==1;
-       x = zscore(x);
-    end
-    
-    nn.testing = 1;
-    nn = nnff(nn, x, y);
-    nn.testing = 0;
-    
-    [~, i] = max(nn.a{end},[],2);
-    [~, g] = max(y,[],2);
-    bad = find(i ~= g);    
+    labels = nnpredict(nn, x);
+    [~, expected] = max(y,[],2);
+    bad = find(labels ~= expected);    
     er = numel(bad) / size(x, 1);
 end

NN/nntrain.m

@@ -10,7 +10,10 @@
     m = size(x, 1);
 
     if nn.normalize_input==1
-       x = zscore(x);
+        [x, mu, sigma] = zscore(x);
+        nn.normalizeMean = mu;
+        sigma(sigma==0) = 1;
+        nn.normalizeStd  = sigma;
     end
 
     batchsize = opts.batchsize;

README.md

@@ -247,7 +247,14 @@ opts.batchsize = 100;  %  Take a mean gradient step over this many samples
 
 assert(er < 0.08, 'Too big error');
 
-
+% Make an artificial one and verify that we can predict it
+x = zeros(1,28,28);
+x(:, 14:15, 6:22) = 1;
+x = reshape(x,1,28^2);
+figure; visualize(x');
+predicted = nnpredict(nn,x)-1;
+
+assert(predicted == 1);
 %% ex2 neural net with L2 weight decay
 rng(0);
 nn = nnsetup([784 100 10]);

tests/test_example_NN.m

@@ -17,7 +17,14 @@
 
 assert(er < 0.08, 'Too big error');
 
-
+% Make an artificial one and verify that we can predict it
+x = zeros(1,28,28);
+x(:, 14:15, 6:22) = 1;
+x = reshape(x,1,28^2);
+figure; visualize(x');
+predicted = nnpredict(nn,x)-1;
+
+assert(predicted == 1);
 %% ex2 neural net with L2 weight decay
 rng(0);
 nn = nnsetup([784 100 10]);