Saving FeedSorter + train() method to train neural network on previously saved feedback

10 files changed, 174 insertions, 64 deletions

diff --git a/app/src/main/java/org/rssin/neurons/FeedSorter.java b/app/src/main/java/org/rssin/neurons/FeedSorter.java
index 550d9e5..157a8f7 100755
--- a/app/src/main/java/org/rssin/neurons/FeedSorter.java
+++ b/app/src/main/java/org/rssin/neurons/FeedSorter.java
@@ -5,7 +5,9 @@ import android.gesture.Prediction;
 import org.rssin.rss.FeedItem;

 

 import java.io.IOException;

+import java.io.Serializable;

 import java.util.ArrayList;

+import java.util.Arrays;

 import java.util.Calendar;

 import java.util.Collection;

 import java.util.Collections;

@@ -18,39 +20,43 @@ import java.util.TimeZone;
 /**

  * Created by Jos on 14-5-2015.

  */

-public class FeedSorter {

+public class FeedSorter implements Serializable{

+    private static final long serialVersionUID = 0;

+

+    private final int MAX_TRAINING_HISTORY = 250;

     private final int SECONDS_IN_DAY = 24 * 60 * 60;

 

     private MultiNeuralNetwork nn = new MultiNeuralNetwork(25, 50);

 

+    private List<TrainingCase> trainingCases = new ArrayList<>();

+

     private int[] isNthMonthInput = new int[12];

     private int[] isNthWeekDayInput = new int[7];

     private int isMorning, isAfternoon, isEvening, isNight, biasInput;

     private Hashtable<String, Integer> categoryInputs = new Hashtable<String, Integer>();

-    private Hashtable<String, Integer> wordInputs = new Hashtable<String, Integer>();

-    private Hashtable<String, Integer> feedSourceInputs = new Hashtable<String, Integer>();

+    //private Hashtable<String, Integer> wordInputs = new Hashtable<String, Integer>();

+    //private Hashtable<String, Integer> feedSourceInputs = new Hashtable<String, Integer>();

 

     public FeedSorter() {

-        //TODO: Load Neural Network

         createNewNetwork();

     }

 

     private void createNewNetwork() {

         biasInput = nn.addInput();

-//        for(int i = 0; i < 12; i++)

-//        {

-//            isNthMonthInput[i] = nn.addInput();

-//        }

-//

-//        for(int i = 0; i < 7; i++)

-//        {

-//            isNthWeekDayInput[i] = nn.addInput();

-//        }

-//

-//        isMorning = nn.addInput();

-//        isAfternoon = nn.addInput();

-//        isEvening = nn.addInput();

-//        isNight = nn.addInput();

+        for(int i = 0; i < 12; i++)

+        {

+            isNthMonthInput[i] = nn.addInput();

+        }

+

+        for(int i = 0; i < 7; i++)

+        {

+            isNthWeekDayInput[i] = nn.addInput();

+        }

+

+        isMorning = nn.addInput();

+        isAfternoon = nn.addInput();

+        isEvening = nn.addInput();

+        isNight = nn.addInput();

     }

 

     private PredictionInterface getPrediction(FeedItem item) {

@@ -75,39 +81,39 @@ public class FeedSorter {
         }

 

         //Set month

-//        Calendar cal = Calendar.getInstance(TimeZone.getTimeZone("UTC"));

-//        for(int i = 0; i < isNthMonthInput.length; i++)

-//        {

-//            if(cal.get(Calendar.MONTH) - cal.getMinimum(Calendar.MONTH) == i)

-//            {

-//                inputs[isNthMonthInput[i]] = 1;

-//            }

-//        }

+        Calendar cal = Calendar.getInstance(TimeZone.getTimeZone("UTC"));

+        for(int i = 0; i < isNthMonthInput.length; i++)

+        {

+            if(cal.get(Calendar.MONTH) - cal.getMinimum(Calendar.MONTH) == i)

+            {

+                inputs[isNthMonthInput[i]] = 1;

+            }

+        }

 

         //Set weekday

-//        for(int i = 0; i < isNthWeekDayInput.length; i++)

-//        {

-//            if(cal.get(Calendar.DAY_OF_WEEK) - cal.getMinimum(Calendar.DAY_OF_WEEK) == i)

-//            {

-//                inputs[isNthMonthInput[i]] = 1;

-//            }

-//        }

+        for(int i = 0; i < isNthWeekDayInput.length; i++)

+        {

+            if(cal.get(Calendar.DAY_OF_WEEK) - cal.getMinimum(Calendar.DAY_OF_WEEK) == i)

+            {

+                inputs[isNthMonthInput[i]] = 1;

+            }

+        }

 

         //Set day

-//        int hourOfDay = cal.get(Calendar.HOUR_OF_DAY);

-//        if(hourOfDay > 6 && hourOfDay < 12)

-//        {

-//            inputs[isMorning] = 1;

-//        }else if(hourOfDay >= 12 && hourOfDay <= 6)

-//        {

-//            inputs[isAfternoon] = 1;

-//        }else if(hourOfDay >= 6 && hourOfDay < 23)

-//        {

-//            inputs[isEvening] = 1;

-//        }else if(hourOfDay >= 23 || hourOfDay <= 6)

-//        {

-//            inputs[isNight] = 1;

-//        }

+        int hourOfDay = cal.get(Calendar.HOUR_OF_DAY);

+        if(hourOfDay > 6 && hourOfDay < 12)

+        {

+            inputs[isMorning] = 1;

+        }else if(hourOfDay >= 12 && hourOfDay <= 6)

+        {

+            inputs[isAfternoon] = 1;

+        }else if(hourOfDay >= 6 && hourOfDay < 23)

+        {

+            inputs[isEvening] = 1;

+        }else if(hourOfDay >= 23 || hourOfDay <= 6)

+        {

+            inputs[isNight] = 1;

+        }

 

         for(String category : item.getCategory())

         {

@@ -126,16 +132,31 @@ public class FeedSorter {
     public void feedback(FeedItem item, Feedback feedback)

     {

         PredictionInterface prediction = getPrediction(item);

-        switch(feedback)

+        prediction.learn(feedback.toExpectedOutput());

+        trainingCases.add(new TrainingCase(prediction.getInputs(), feedback));

+

+        while(trainingCases.size() > MAX_TRAINING_HISTORY)

+        {

+            trainingCases.remove(0);

+        }

+    }

+

+    /**

+     * Runs an iteration of training, using feedback that was provided previously using FeedSorter.feedback(...).

+     */

+    public void train()

+    {

+        for(TrainingCase t : trainingCases)

         {

-            case Like:

-                prediction.learn(1);

-                break;

-            case Dislike:

-                prediction.learn(-1);

-                break;

-            default:

-                throw new IllegalArgumentException("feedback");

+            double[] inputs = t.getInputs();

+            if(inputs.length < nn.getInputCount())

+            {

+                // Resize array to fit new input size

+                inputs = Arrays.copyOf(inputs, nn.getInputCount());

+            }

+

+            PredictionInterface prediction = nn.computeOutput(inputs);

+            prediction.learn(t.getFeedback().toExpectedOutput());

         }

     }

 

diff --git a/app/src/main/java/org/rssin/neurons/FeedSorterStorage.java b/app/src/main/java/org/rssin/neurons/FeedSorterStorage.java
new file mode 100755
index 0000000..1479a3b
--- /dev/null
+++ b/app/src/main/java/org/rssin/neurons/FeedSorterStorage.java
@@ -0,0 +1,44 @@
+package org.rssin.neurons;

+

+import android.content.Context;

+

+import java.io.File;

+import java.io.FileNotFoundException;

+import java.io.IOException;

+import java.io.ObjectInputStream;

+import java.io.ObjectOutputStream;

+

+/**

+ * Created by Jos on 20-5-2015.

+ */

+public class FeedSorterStorage {

+    private final String file = "ml.dat";

+    public FeedSorterStorage()

+    {}

+

+    /**

+     * Reads a FeedSorter from an internal file.

+     * @param context The context

+     * @return The loaded FeedSorter

+     * @throws IOException

+     * @throws ClassNotFoundException

+     */

+    public FeedSorter loadSorter(Context context) throws IOException, ClassNotFoundException {

+        ObjectInputStream is = new ObjectInputStream(context.openFileInput(file));

+        FeedSorter read = (FeedSorter)is.readObject();

+        is.close();

+        return read;

+    }

+

+    /**

+     * Saves the FeedSorter to an internal file.

+     * @param context The context

+     * @param s The FeedSorter to save

+     * @throws IOException

+     */

+    public void saveSorter(Context context, FeedSorter s) throws IOException {

+        ObjectOutputStream os = new ObjectOutputStream(context.openFileOutput(file, Context.MODE_PRIVATE));

+        os.writeObject(s);

+        os.close();

+    }

+}

diff --git a/app/src/main/java/org/rssin/neurons/Feedback.java b/app/src/main/java/org/rssin/neurons/Feedback.java
index af5f1b3..9652b2b 100755
--- a/app/src/main/java/org/rssin/neurons/Feedback.java
+++ b/app/src/main/java/org/rssin/neurons/Feedback.java
@@ -4,6 +4,18 @@ package org.rssin.neurons;
  * Created by Jos on 19-5-2015.

  */

 public enum Feedback {

-    Like,

-    Dislike

+    Like, Dislike;

+

+    double toExpectedOutput()

+    {

+        switch(this)

+        {

+            case Like:

+                return 1;

+            case Dislike:

+                return -1;

+            default:

+                throw new IllegalArgumentException();

+        }

+    }

 }

diff --git a/app/src/main/java/org/rssin/neurons/MultiNeuralNetwork.java b/app/src/main/java/org/rssin/neurons/MultiNeuralNetwork.java
index 50fa707..09dfc21 100755
--- a/app/src/main/java/org/rssin/neurons/MultiNeuralNetwork.java
+++ b/app/src/main/java/org/rssin/neurons/MultiNeuralNetwork.java
@@ -4,6 +4,7 @@ package org.rssin.neurons;
  * Created by Jos on 14-5-2015.

  */

 class MultiNeuralNetwork {

+    private static final long serialVersionUID = 0;

     private NeuralNetwork[] networks;

 

     public MultiNeuralNetwork(int numNetworks, int numHiddenNodes) {

diff --git a/app/src/main/java/org/rssin/neurons/MultiNeuralNetworkPrediction.java b/app/src/main/java/org/rssin/neurons/MultiNeuralNetworkPrediction.java
index 8df99e5..dc85261 100755
--- a/app/src/main/java/org/rssin/neurons/MultiNeuralNetworkPrediction.java
+++ b/app/src/main/java/org/rssin/neurons/MultiNeuralNetworkPrediction.java
@@ -7,6 +7,11 @@ class MultiNeuralNetworkPrediction implements PredictionInterface {
     private PredictionInterface[] predictions;

     MultiNeuralNetworkPrediction(PredictionInterface[] predictions)

     {

+        if(predictions.length <= 0)

+        {

+            throw new IllegalArgumentException("predictions");

+        }

+

         this.predictions = predictions;

     }

 

@@ -27,4 +32,9 @@ class MultiNeuralNetworkPrediction implements PredictionInterface {
             prediction.learn(expectedOutput);

         }

     }

+

+    public double[] getInputs()

+    {

+        return predictions[0].getInputs();

+    }

 }

diff --git a/app/src/main/java/org/rssin/neurons/NeuralNetwork.java b/app/src/main/java/org/rssin/neurons/NeuralNetwork.java
index f5ee569..3766262 100755
--- a/app/src/main/java/org/rssin/neurons/NeuralNetwork.java
+++ b/app/src/main/java/org/rssin/neurons/NeuralNetwork.java
@@ -4,6 +4,7 @@ package org.rssin.neurons;
  * Created by Jos on 14-5-2015.

  */

 class NeuralNetwork {

+    private static final long serialVersionUID = 0;

     private Neuron[] hiddenNodes;

     private Neuron outputNode;

 

@@ -25,11 +26,7 @@ class NeuralNetwork {
     public int addInput() {

         int result = 0;

         for (int i = 0; i < hiddenNodes.length; i++) {

-<<<<<<< HEAD
             result = hiddenNodes[i].addWeight();

-=======
-            //result = hiddenNodes[i].AddWeight();

->>>>>>> 7336ef600f6bd472b9aa1d59ad6418ff5c543044
         }

 

         return result;

diff --git a/app/src/main/java/org/rssin/neurons/NeuralNetworkPrediction.java b/app/src/main/java/org/rssin/neurons/NeuralNetworkPrediction.java
index 162ba21..9d6fc89 100755
--- a/app/src/main/java/org/rssin/neurons/NeuralNetworkPrediction.java
+++ b/app/src/main/java/org/rssin/neurons/NeuralNetworkPrediction.java
@@ -16,11 +16,11 @@ class NeuralNetworkPrediction implements PredictionInterface {
         this.nn = nn;

     }

 

-    double[] getInputs() {

+    public double[] getInputs() {

         return inputs;

     }

 

-    double[] getIntermediateValues() {

+    public double[] getIntermediateValues() {

         return intermediateValues;

     }

 

diff --git a/app/src/main/java/org/rssin/neurons/Neuron.java b/app/src/main/java/org/rssin/neurons/Neuron.java
index 5df9d47..724732d 100755
--- a/app/src/main/java/org/rssin/neurons/Neuron.java
+++ b/app/src/main/java/org/rssin/neurons/Neuron.java
@@ -8,6 +8,7 @@ import java.util.Random;
  * Created by Jos on 14-5-2015.

  */

 class Neuron {

+    private static final long serialVersionUID = 0;

     private static Random r = new Random();

 

     private List<Double> weights = new ArrayList<Double>();

diff --git a/app/src/main/java/org/rssin/neurons/PredictionInterface.java b/app/src/main/java/org/rssin/neurons/PredictionInterface.java
index e130f4d..27a214f 100755
--- a/app/src/main/java/org/rssin/neurons/PredictionInterface.java
+++ b/app/src/main/java/org/rssin/neurons/PredictionInterface.java
@@ -6,4 +6,5 @@ package org.rssin.neurons;
 interface PredictionInterface {

     public double getOutput();

     public void learn(double expectedOutput);

+    public double[] getInputs();

 }

diff --git a/app/src/main/java/org/rssin/neurons/TrainingCase.java b/app/src/main/java/org/rssin/neurons/TrainingCase.java
new file mode 100755
index 0000000..e2680f8
--- /dev/null
+++ b/app/src/main/java/org/rssin/neurons/TrainingCase.java
@@ -0,0 +1,23 @@
+package org.rssin.neurons;

+

+/**

+ * Created by Jos on 20-5-2015.

+ */

+class TrainingCase {

+    private double[] inputs;

+    private Feedback feedback;

+

+    public TrainingCase(double[] inputs, Feedback feedback)

+    {

+        this.inputs = inputs;

+        this.feedback = feedback;

+    }

+

+    public double[] getInputs() {

+        return inputs;

+    }

+

+    public Feedback getFeedback() {

+        return feedback;

+    }

+}