Fix forward pass issue with MLN.score(DataSet)

jungleTail · Mar 16, 2016 · e9d7a0a · e9d7a0a
1 parent 56f9a4d
commit e9d7a0a
Show file tree

Hide file tree

Showing 2 changed files with 26 additions and 1 deletion.
diff --git a/deeplearning4j-core/src/main/java/org/deeplearning4j/nn/multilayer/MultiLayerNetwork.java b/deeplearning4j-core/src/main/java/org/deeplearning4j/nn/multilayer/MultiLayerNetwork.java
@@ -1629,10 +1629,11 @@ public double score(DataSet data, boolean training){
         boolean hasMaskArray = data.hasMaskArrays();
         if(hasMaskArray) setLayerMaskArrays(data.getFeaturesMaskArray(),data.getLabelsMaskArray());
         // activation for output layer is calculated in computeScore
-        feedForwardToLayer(layers.length - 2, data.getFeatureMatrix(),training);
+        List<INDArray> activations = feedForwardToLayer(layers.length - 2, data.getFeatureMatrix(),training);
         setLabels(data.getLabels());
         if( getOutputLayer() instanceof BaseOutputLayer ){
             BaseOutputLayer<?> ol = (BaseOutputLayer<?>)getOutputLayer();
+            ol.setInput(activations.get(activations.size()-1));     //Feedforward doesn't include output layer for efficiency
             ol.setLabels(data.getLabels());
             ol.computeScore(calcL1(),calcL2(), training);
             this.score = ol.score();

diff --git a/deeplearning4j-core/src/test/java/org/deeplearning4j/nn/multilayer/MultiLayerTest.java b/deeplearning4j-core/src/test/java/org/deeplearning4j/nn/multilayer/MultiLayerTest.java
@@ -631,6 +631,30 @@ public void testScoreExamples(){
         }
     }
 
+    @Test
+    public void testDataSetScore(){
+
+        Nd4j.getRandom().setSeed(12345);
+        MultiLayerConfiguration conf = new NeuralNetConfiguration.Builder()
+                .regularization(false)
+                .learningRate(1.0)
+                .weightInit(WeightInit.XAVIER)
+                .seed(12345L)
+                .list()
+                .layer(0, new DenseLayer.Builder().nIn(4).nOut(3).activation("sigmoid").build())
+                .layer(1, new OutputLayer.Builder(LossFunctions.LossFunction.MCXENT).activation("softmax").nIn(3).nOut(3).build())
+                .pretrain(false).backprop(true)
+                .build();
+
+        MultiLayerNetwork net = new MultiLayerNetwork(conf);
+        net.init();
+
+        INDArray in = Nd4j.create(new double[]{1.0,2.0,3.0,4.0});
+        INDArray out = Nd4j.create(new double[]{1,0,0});
+
+        double score = net.score(new DataSet(in,out));
+    }
+
     @Test
     @Ignore
     public void testCid() throws Exception {