vol3:java:more refactoring, now ready to address: #31

vol3/vol3-java-examples/src/main/java/com/heatonresearch/aifh/ann/BasicLayer.java

@@ -49,7 +49,14 @@ public class BasicLayer extends WeightedLayer {
      */
     private boolean hasBias;
 
+    /**
+     * The output from each of the layers.
+     */
     private FlatVolume layerOutput;
+
+    /**
+     * The sums from each of the layers, esseitnally the output prior to activation function.
+     */
     private FlatVolume layerSums;
 
     /**
@@ -88,9 +95,6 @@ public BasicLayer(final ActivationFunction theActivation, boolean theHasBias, in
         this(theActivation,theHasBias,new int[] {theCount,1,1});
     }
 
-
-
-
     /**
      * @return the count
      */
@@ -145,43 +149,14 @@ public int[] getDimensionCounts() {
      * {@inheritDoc}
      */
     @Override
-    public int getWeightDepthUnit() {
-        Layer previousLayer = getOwner().getPreviousLayer(this);
-        int prevCount;
-        if( previousLayer instanceof Conv2DLayer ) {
-            prevCount = (((Conv2DLayer)previousLayer).getFilterColumns() *
-                    ((Conv2DLayer)previousLayer).getFilterRows());
-        } else {
-            if( previousLayer.getDimensionCounts().length==1) {
-                prevCount = previousLayer.getCount();
-            } else {
-                prevCount = previousLayer.getDimensionCounts()[0] * previousLayer.getDimensionCounts()[1];
-            }
-        }
-        if(previousLayer.hasBias()) {
-            prevCount++;
-        }
-
-
-        return prevCount * getNeuronDepthUnit();
+    public FlatVolume getLayerOutput() {
+        return layerOutput;
     }
 
     /**
      * {@inheritDoc}
      */
     @Override
-    public int getNeuronDepthUnit() {
-        if( this.count.length==3) {
-            return this.count[0] * this.count[1];
-        } else {
-            return this.count[0];
-        }
-    }
-
-    public FlatVolume getLayerOutput() {
-        return layerOutput;
-    }
-
     public FlatVolume getLayerSums() {
         return layerSums;
     }

vol3/vol3-java-examples/src/main/java/com/heatonresearch/aifh/ann/BasicNetwork.java

@@ -69,6 +69,9 @@ public class BasicNetwork implements RegressionAlgorithm, ClassificationAlgorith
      */
     private boolean networkTraining;
 
+    /**
+     * The outputs/activations from each layer.
+     */
     private FlatData layerOutput = new FlatData();
 
     /**
@@ -100,25 +103,6 @@ public void compute(final double[] input, final double[] output) {
                 output, 0, this.outputCount);
     }
 
-    /**
-     * @return The total number of neurons in the neural network.
-     */
-    public int getNeuronCount() {
-        int result = 0;
-        for(Layer layer: this.layers) {
-            result+=layer.getTotalCount();
-        }
-        return result;
-    }
-
-    /**
-     * @return The length of the array the network would encode to.
-     */
-    public int getEncodeLength() {
-        return this.weights.getLength();
-    }
-
-
     /**
      * @return The number of input neurons.
      */
@@ -149,16 +133,6 @@ public void setInputCount(final int inputCount) {
         this.inputCount = inputCount;
     }
 
-
-
-    /**
-     * Set the output count.
-     * @param outputCount The output count.
-     */
-    public void setOutputCount(final int outputCount) {
-        this.outputCount = outputCount;
-    }
-
     /**
      * Get the weight between the two layers.
      * @param fromLayer The from layer.
@@ -226,11 +200,9 @@ public void finalizeStructure() {
         this.inputCount = this.layers.get(0).getCount();
         this.outputCount = this.layers.get(layerCount - 1).getCount();
 
-        TempStructureCounts counts = new TempStructureCounts();
-
         for (int i = this.layers.size() - 1; i >= 0; i--) {
             final Layer layer = this.layers.get(i);
-            layer.finalizeStructure(this, i, counts);
+            layer.finalizeStructure(this, i);
             this.layerOutput.addFlatObject(layer.getLayerSums());
             this.layerOutput.addFlatObject(layer.getLayerOutput());
             if( layer.getWeightMatrix()!=null ) {
@@ -279,14 +251,7 @@ public void setWeight(final int fromLayer, final int fromNeuron,
                             + fromLayer);
         }
 
-        final int weightBaseIndex
-                = this.layers.get(fromLayer+1).getWeightIndex();
-        final int count = this.layers.get(fromLayer).getTotalCount();
-
-        final int weightIndex = weightBaseIndex + fromNeuron
-                + (toNeuron * count);
-
-        getWeights()[weightIndex] = value;
+        this.layers.get(fromLayer+1).getWeightMatrix().set(toNeuron,fromNeuron, value);
     }
 
     /**
@@ -388,24 +353,24 @@ public void setNetworkTraining(boolean networkTraining) {
         this.networkTraining = networkTraining;
     }
 
+    /**
+     * @return The input layer.
+     */
     public Layer getInputLayer() {
         return this.layers.get(0);
     }
 
+    /**
+     * @return The output layer.
+     */
     public Layer getOutputLayer() {
         return this.layers.get(this.layers.size()-1);
     }
 
+    /**
+     * @return The output from each of the layers.
+     */
     public FlatData getLayerOutput() {
         return layerOutput;
     }
-
-    public void dumpOutputs() {
-        int i = 0;
-        for(Layer layer: this.layers) {
-            i++;
-            System.out.println("Layer #" + i + ":Sums=" + layer.getLayerSums()
-                    + ",Output=" + layer.getLayerOutput()  );
-        }
-    }
 }

vol3/vol3-java-examples/src/main/java/com/heatonresearch/aifh/ann/Conv2DLayer.java

@@ -109,8 +109,8 @@ public Conv2DLayer(final ActivationFunction theActivation, int theNumFilters, in
      * {@inheritDoc}
      */
     @Override
-    public void finalizeStructure(BasicNetwork theOwner, int theLayerIndex, TempStructureCounts counts) {
-        super.finalizeStructure(theOwner,theLayerIndex,counts);
+    public void finalizeStructure(BasicNetwork theOwner, int theLayerIndex) {
+        super.finalizeStructure(theOwner,theLayerIndex);
 
         Layer prevLayer = (getLayerIndex()>0) ? getOwner().getLayers().get(getLayerIndex()-1) : null;
         Layer nextLayer = (getLayerIndex()<getOwner().getLayers().size()-1) ? getOwner().getLayers().get(getLayerIndex()+1) : null;
@@ -127,23 +127,6 @@ public void finalizeStructure(BasicNetwork theOwner, int theLayerIndex, TempStru
         this.outRows = Math.floor((inRows + this.padding * 2 - this.filterColumns) / this.stride + 1);
     }
 
-    /**
-     * {@inheritDoc}
-     */
-    @Override
-    public int getWeightDepthUnit() {
-        Layer previousLayer = getOwner().getPreviousLayer(this);
-        return previousLayer.getNeuronDepthUnit() * getNeuronDepthUnit();
-    }
-
-    /**
-     * {@inheritDoc}
-     */
-    @Override
-    public int getNeuronDepthUnit() {
-        return this.filterColumns * this.filterRows;
-    }
-
     @Override
     public FlatVolume getLayerOutput() {
         return this.layerOutput;

vol3/vol3-java-examples/src/main/java/com/heatonresearch/aifh/ann/Layer.java

@@ -58,10 +58,8 @@ public interface Layer {
      * Finalize the structure of this layer.
      * @param theOwner The neural network that owns this layer.
      * @param theLayerIndex The zero-based index of this layer.
-     * @param counts The counts structure to track the weight and neuron counts.
      */
-    void finalizeStructure(BasicNetwork theOwner, int theLayerIndex,
-                           TempStructureCounts counts);
+    void finalizeStructure(BasicNetwork theOwner, int theLayerIndex);
 
     /**
      * Compute this layer.
@@ -74,16 +72,6 @@ void finalizeStructure(BasicNetwork theOwner, int theLayerIndex,
      */
     void computeGradient(GradientCalc calc);
 
-    /**
-     * @return The start of this layer's weights in the weight vector.
-     */
-    int getWeightIndex();
-
-    /**
-     * @return The start of this layer's neurons in the neuron vector.
-     */
-    int getNeuronIndex();
-
     /**
      * @return This layer's index in the layer stack.
      */
@@ -119,24 +107,18 @@ void finalizeStructure(BasicNetwork theOwner, int theLayerIndex,
     int[] getDimensionCounts();
 
     /**
-     * Get the number of weights in a single unit.  For  non-convolution layer, this is the total number of weights.
-     * For a convolution network, this is the number of weights per filter.
-     * @return The weights per depth unit.
+     * @return The output from each of the layers.
      */
-    int getWeightDepthUnit();
+    FlatVolume getLayerOutput();
 
     /**
-     * Get the number of neurons in a single unit.  For non-convolution layers, this is the total number of neurons in
-     * this layer.  For convolution networks this is the number of neurons per filter.
-     * @return The neurons per depth unit.
+     * @return The sums from each of the layers.
      */
-    int getNeuronDepthUnit();
-
-    FlatVolume getLayerOutput();
-
     FlatVolume getLayerSums();
 
+    /**
+     * @return The weight matrix.
+     */
     FlatMatrix getWeightMatrix();
 
-
 }

vol3/vol3-java-examples/src/main/java/com/heatonresearch/aifh/ann/MaxPoolLayer.java

@@ -75,7 +75,7 @@ public ActivationFunction getActivation() {
      * {@inheritDoc}
      */
     @Override
-    public void finalizeStructure(BasicNetwork theOwner, int theLayerIndex, TempStructureCounts counts) {
+    public void finalizeStructure(BasicNetwork theOwner, int theLayerIndex) {
 
     }
 
@@ -90,16 +90,6 @@ public void computeGradient(GradientCalc calc) {
 
     }
 
-    @Override
-    public int getWeightIndex() {
-        return 0;
-    }
-
-    @Override
-    public int getNeuronIndex() {
-        return 0;
-    }
-
 
     @Override
     public int getLayerIndex() { return 0; }
@@ -129,16 +119,6 @@ public int[] getDimensionCounts() {
         return new int[0];
     }
 
-    @Override
-    public int getWeightDepthUnit() {
-        return 0;
-    }
-
-    @Override
-    public int getNeuronDepthUnit() {
-        return 0;
-    }
-
     @Override
     public FlatVolume getLayerOutput() {
         return null;