reimplemented neural networks with a better model, need to fix lag

core/src/logic/neural/Brain.java

@@ -10,75 +10,106 @@ import java.util.ArrayList;
 public class Brain {
 
     public static final float bias = 0.5f;
-    private ArrayList<Neuron> inputs, outputs, hidden;
+    private Neuron[][] neurons;
+    private int nInputs;
 
     public Brain(int nInputs, int nOutputs, int hiddenLayers, int neuronsPerHiddenLayer) {
-        inputs = new ArrayList<Neuron>(nInputs);
+        this.nInputs = nInputs;
-        outputs = new ArrayList<Neuron>(nOutputs);
+        neurons = new Neuron[hiddenLayers + 2][Math.max(nInputs, Math.max(nOutputs, neuronsPerHiddenLayer))];
-        hidden = new ArrayList<Neuron>(hiddenLayers * neuronsPerHiddenLayer);
         populate(nInputs, nOutputs, hiddenLayers, neuronsPerHiddenLayer);
     }
 
     private void populate(int nInputs, int nOutputs, int hiddenLayers, int neuronsPerHiddenLayer) {
         // Create input neurons
         for (int i = 0; i < nInputs; i++) {
-            inputs.add(new Neuron(0,bias));
+            neurons[0][i] = new Neuron(0, bias, this);
+            Log.log(Log.DEBUG, "Adding Input Layer Neuron " + (i + 1));
         }
         // popiulate hidden layers
         for (int i = 0; i < hiddenLayers; i++) {
             for (int j = 0; j < neuronsPerHiddenLayer; j++) {
                 // create neuron
-                Neuron n = new Neuron(i + 1,bias);
+                Neuron n = new Neuron(i + 1, bias, this);
-                // add connections
+                neurons[i + 1][j] = n;
-                for (Neuron s : inputs) {
+                Log.log(Log.DEBUG, "Adding Hidden Layer " + (i + 1) + " Neuron " + (j + 1));
-                    n.getInputs().add(new NeuralConnection(randWeight(), s));
-                }
-                hidden.add(n);
-                Log.log(Log.DEBUG,"Adding Hidden Layer " + (i + 1) + " Neuron " + j + " with " + inputs.size() + " inputs");
             }
         }
         // populate output layer
         for (int i = 0; i < nOutputs; i++) {
             // add neuron
-            Neuron n = new Neuron(hiddenLayers + 1,bias);
+            Neuron n = new Neuron(hiddenLayers + 1, bias, this);
-            int conn = 0;
+            neurons[hiddenLayers + 1][i] = n;
-            for (Neuron s : hidden) {
+            Log.log(Log.DEBUG, "Adding Output Layer Neuron " + (i + 1));
-                // add connections where applicable
-                if (s.getLayer() == hiddenLayers) {
-                    conn++;
-                    n.getInputs().add(new NeuralConnection(randWeight(), s));
-                }
-            }
-            Log.log(Log.DEBUG,"Adding Output Layer Neuron " + i + " with " + conn + " inputs");
-            outputs.add(n);
         }
     }
 
     private float randWeight() {
-        return (float) Math.random()*5 - 2.5f;
+        return (float) Math.random() * 5 - 2.5f;
     }
 
     public void input(float[] values) {
         for (int i = 0; i < values.length; i++) {
-            inputs.get(i).setOutput(values[i]);
+            neurons[0][i].setOutput(values[i]);
         }
     }
 
     public float[] compute() {
-        for (Neuron n : hidden) {
+        float[] res = new float[neurons[neurons.length - 1].length];
-            n.clearCachedValue();
+        for (int i = 0; i < neurons[neurons.length - 1].length; i++) {
-        }
+            Neuron n = neurons[neurons.length - 1][i];
-        float[] res = new float[outputs.size()];
+            if (n != null) {
-        for (int i = 0; i < outputs.size(); i++) {
+                res[i] = n.compute();
-            Neuron n = outputs.get(i);
+            }
-            n.clearCachedValue();
-            res[i] = n.compute();
         }
         return res;
     }
 
-    public void mutate(float mutationFactor) {
+    public void map(float[][][] map) {
-        for(Neuron n : hidden) n.mutate(mutationFactor);
+        // Populate with new neurons
+        for (int j = 0; j < map.length; j++) {
+            for (int i = 0; i < map[j].length; i++) {
+                if (map[j] == null || map[i] == null) {
+                    continue;
+                }
+                neurons[j][i] = new Neuron(j, bias, this);
+                neurons[j][i].setWeights(map[j][i]);
+            }
+        }
+    }
+
+    public float[][][] getMap() {
+        float[][][] res = new float[neurons.length][neurons[1].length][neurons[1].length];
+        for (int i = 0; i < neurons.length; i++) // layers
+        {
+            for (int j = 0; i < neurons[i].length; j++) // neurons per layer
+            {
+                if (neurons[i][j] == null) {
+                    continue;
+                }
+                res[i][j] = neurons[i][j].getWeights();
+            }
+        }
+        return res;
+    }
+
+    public float[][][] mutate(float mutationFactor) {
+        float[][][] res = new float[neurons.length][neurons[1].length][neurons[1].length];
+        for (int i = 0; i < neurons.length; i++) // layers
+        {
+            for (int j = 0; i < neurons[i].length; j++) // neurons per layer
+            {
+                res[i][j] = neurons[i][j].mutate(mutationFactor);
+            }
+        }
+        return res;
+    }
+
+    public Neuron[][] getNeurons() {
+        return neurons;
+    }
+
+    public int howManyInputNeurons() {
+        return nInputs;
     }
 
 }

core/src/logic/neural/NeuralConnection.java

@@ -1,50 +0,0 @@
-/*
- * To change this license header, choose License Headers in Project Properties.
- * To change this template file, choose Tools | Templates
- * and open the template in the editor.
- */
-package logic.neural;
-
-/**
- *
- * @author fazo
- */
-public class NeuralConnection {
-
-    private float weight = 1;
-
-    private final Neuron source;
-    private float cachedValue;
-    private boolean cachedValueValid = false;
-
-    public NeuralConnection(float weight, Neuron source) {
-        this.source = source;
-        this.weight = weight;
-    }
-
-    public float compute() {
-        if (cachedValueValid) {
-            return cachedValue;
-        }
-        // get value from Neuron
-        cachedValueValid = true;
-        return cachedValue = source.compute() * getWeight();
-    }
-
-    public void mutate(float mutationFactor) {
-        float mutation = (float) (Math.random() * mutationFactor - mutationFactor/2);
-        weight += mutation;
-    }
-    
-    public void clearCachedValue() {
-        cachedValueValid = false;
-    }
-
-    public float getWeight() {
-        return weight;
-    }
-
-    public void setWeight(float weight) {
-        this.weight = weight;
-    }
-}

core/src/logic/neural/Neuron.java

@@ -7,6 +7,9 @@ package logic.neural;
 
 import com.mygdx.game.Log;
 import java.util.ArrayList;
+import java.util.Arrays;
+import java.util.logging.Level;
+import java.util.logging.Logger;
 
 /**
  *
@@ -14,38 +17,65 @@ import java.util.ArrayList;
  */
 public class Neuron {
 
-    private ArrayList<NeuralConnection> inputs;
+    private float[] weights;
     private float bias, output;
     private boolean isInputNeuron;
     private int layer;
-    private float cachedValue;
+    private Brain brain;
-    private boolean cachedValueValid = false;
 
-    public Neuron(int layer, float bias) {
+    public Neuron(int layer, float bias, Brain brain) {
+        this(layer, bias, brain, null);
+    }
+
+    public Neuron(int layer, float bias, Brain brain, float[] weights) {
+        this.brain = brain;
         this.layer = layer;
-        inputs = new ArrayList<NeuralConnection>();
+        if (weights == null) {
+            scramble();
+        } else {
+            this.weights = weights;
+        }
+    }
+
+    private void scramble() {
+        // init weights
+        if (layer > 1) {
+            weights = new float[brain.getNeurons()[layer - 1].length];
+        } else if (layer == 1) {
+            weights = new float[brain.howManyInputNeurons()];
+        } else { // layer 0
+            isInputNeuron = true;
+            weights = new float[0];
+        }
+        // Put random weights
+        for (int i = 0; i < weights.length; i++) {
+            weights[i] = (float) (Math.random() * 5 - 2.5f);
+        }
     }
 
     public float compute() {
         if (isInputNeuron) {
             return output;
         }
-        if (cachedValueValid) {
-            return cachedValue;
-        }
         float a = bias * -1; // activation
-        for (NeuralConnection i : inputs) {
+        for (int i = 0; i < weights.length; i++) {
-            a += i.compute();
+            //if(brain == null) System.out.println("BRAINS NULL"); else if(brain.getNeurons() == null) System.out.println("NEURONS NULL");
+            //System.out.println(Arrays.toString(brain.getNeurons()));
+            Neuron n = brain.getNeurons()[layer - 1][i];
+            a += n.compute() * weights[i];
         }
-        cachedValueValid = true;
         // sigmoid function
-        cachedValue = (float) (1 / (1 + Math.pow(Math.E, a * -1)));
+        float res = (float) (1 / (1 + Math.pow(Math.E, a * -1)));
-        Log.log(Log.DEBUG,"Computed Value "+cachedValue+" for neuron");
+        Log.log(Log.DEBUG, "Computed Value " + res + " for neuron");
-        return cachedValue;
+        return res;
     }
 
-    public void mutate(float mutationFactor){
+    public float[] mutate(float mutationFactor) {
-        for(NeuralConnection n : inputs) n.mutate(mutationFactor);
+        float[] mutatedWeights = new float[weights.length];
+        for (int i = 0; i < weights.length; i++) {
+            mutatedWeights[i] = weights[i] + mutationFactor - mutationFactor / 2;
+        }
+        return mutatedWeights;
     }
 
     public void setOutput(float output) {
@@ -53,10 +83,6 @@ public class Neuron {
         this.output = output;
     }
 
-    public ArrayList<NeuralConnection> getInputs() {
-        return inputs;
-    }
-
     public float getBias() {
         return bias;
     }
@@ -77,9 +103,12 @@ public class Neuron {
         this.layer = layer;
     }
 
-    public void clearCachedValue() {
+    public float[] getWeights() {
-        cachedValueValid = false;
+        return weights;
-        for(NeuralConnection n : inputs) n.clearCachedValue();
+    }
+
+    public void setWeights(float[] weights) {
+        this.weights = weights;
     }
 
 }