Make training slightly more robust

server/training.js

@@ -23,7 +23,6 @@ Meteor.startup(function() {
   }
   console.log('Got ' + training_data.length + ' rows of training data.');
 
-  var input = new convnetjs.Vol(1, 1, 8);
   var net = new convnetjs.Net();
   net.makeLayers([
     {type: 'input', out_sx: 1, out_sy: 1, out_depth: 8},
@@ -33,20 +32,24 @@ Meteor.startup(function() {
   ]);
   var trainer = new convnetjs.Trainer(
       net, {method: 'adadelta', l2_decay: 0.001, batch_size: 10});
-  for (var iteration = 0; iteration < 1; iteration++) {
+  var input = new convnetjs.Vol(1, 1, 8);
+  for (var iteration = 0; iteration < 10; iteration++) {
+    var loss = 0;
     for (var i = 0; i < training_data.length; i++) {
       assert(input.w.length === training_data[i][0].length);
       input.w = training_data[i][0];
-      trainer.train(input, training_data[i][1]);
+      var stats = trainer.train(input, [training_data[i][1]]);
+      assert(!isNaN(stats.loss))
+      loss += stats.loss;
     }
-    console.log('Completed iteration:', iteration);
+    console.log('Iteration', iteration, 'loss:', loss/training_data.length);
   }
   console.log('Trained neural network.');
 
   function net_classifier(features) {
     assert(input.w.length === features.length);
     input.w = features;
-    return net.forward(input).w[0];
+    return net.forward(input).w[0] || 0;
   }
   console.log('Neural-net accuracy:', evaluate(glyphs, net_classifier));
 });