Made some code independant by adding another dataset

Author: Aatmaj-Zephyr

.gitignore

@@ -1 +1,5 @@
-/.DS_Store
+/.DS_Store
+/.mnist_train.csv
+/.mnist_train.csv
+/.emnist-letters-test.csv
+/.emnist-letters-train.csv

Main.java

@@ -3,22 +3,24 @@
 public class Main {
 	public static void main(String[] args) {
 		parameter.setOutputFile("output.txt",true);
-		parameter.setLayerArray(784, 32, 16, 16, 10);
-		parameter.setTrainingFileReader("mnist_train.csv", "mnist");
-		parameter.setTestingFileReader("mnist_test.csv", "mnist");
+		parameter.setLayerArray(784, 32, 16, 16, 26);
+		parameter.setTrainingFileReader("emnist-letters-train.csv", "mnist");
+		parameter.setTestingFileReader("emnist-letters-test.csv", "mnist");
+		
 		parameter.setLearningRate(1);
 		parameter.setBiasLearningRate(1);
 		parameter.setEpsillion(0);
 		parameter.setBatchsize(10);
 		parameter.setRectificationFunction("sigmoid");
+
 		Trainer myTrainer = new Trainer();
-		myTrainer.train(50,10);
+		myTrainer.train(88800,1);
 		NeuronObserver myNeuronObserver = new NeuronObserver();
 		myNeuronObserver.setModel(myTrainer.getLayerManager());
 		myTrainer.test(9990);
 
 		myNeuronObserver.addNeuronToBeObserved(1, 31);
-		myTrainer.printConfusionMatrix();
+		//myTrainer.printConfusionMatrix();
 		myTrainer.test(2);
 
 		myNeuronObserver.clear();

ann4j/MNISTDataBaseFileReader.java

@@ -35,6 +35,7 @@ public void next() {
             inputArray = generateInputFromBigArray(array);
 
         } catch (IOException ex) {
+            ex.printStackTrace();
         }
     }
 
@@ -68,7 +69,7 @@ public ArrayList<Double> generateExpectedOutputArrayFromLabel() {
         // this needs to be overridden for changing the dataset
         // responsible for generating the output neurons.
         ArrayList<Double> expectedOutputArray = new ArrayList<Double>();
-        for (int i = 0; i < 10; i++) {
+        for (int i = 0; i < parameter.numberOfOutputNeurons; i++) {
             if (i == label) {
                 // Works only for 0-9 digits, 0 output neuron corresponds to 0, 1st to 1 etc.
                 // But you can change this and jumble the order as required. This may be one of

ann4j/ModelEvaluator.java

@@ -13,7 +13,7 @@ public class ModelEvaluator {
 
     public void updateConfusionMatrix(ArrayList<Double> expectedOutputArray, int predictedNeuronNum) {
         ArrayList<Double> temp = new ArrayList<Double>();
-        for (int i = 0; i <= expectedOutputArray.size(); i++) {
+        for (int i = 0; i <= parameter.numberOfOutputNeurons; i++) {
             if (i == predictedNeuronNum) {
                 temp.add(i, 1.0);
             } else {
@@ -42,10 +42,10 @@ public void updateConfusionMatrix(ArrayList<Double> expectedOutputArray, ArrayLi
         // This is checking if the confusion matrix has been initialized, if it has not
         // been
         // initialized, it initializes it.
-        if (confusionMatrix.size() != expectedOutputArray.size()) {// arraylist has not been initialized
-            initializeList(expectedOutputArray.size());
+        if (confusionMatrix.size() != parameter.numberOfOutputNeurons) {// arraylist has not been initialized
+            initializeList(parameter.numberOfOutputNeurons);
         }
-        for (int i = 0; i < expectedOutputArray.size(); i++) {
+        for (int i = 0; i < parameter.numberOfOutputNeurons; i++) {
 
             // This is updating the confusion matrix.
             if (expectedOutputArray.get(i) == 1.0 & actualOutputArray.get(i) == 1.0) { // True positives

ann4j/Trainer.java

@@ -30,7 +30,6 @@ public LayerManager getLayerManager() {
     public void train(int noOfSamples, int epochs) {
         for(int j = 0; j < epochs; j++){
 
-            parameter.setTrainingFileReader("mnist_train.csv", "mnist");
             this.trainingFileReader = parameter.getTrainingFileReader();
             for (int i = 0; i < noOfSamples; i++) {
 

ann4j/parameter.java

@@ -15,6 +15,9 @@ public class parameter {
     private static int batchsize;
     private static double biasLearningRate;
     private static double epsillion;
+  // Used to store the number of input and output neurons.
+    public static int numberOfOutputNeurons;
+    public static int numberOfInputNeurons;
 
     public static double getBiasLearningRate() {
         return biasLearningRate;
@@ -51,6 +54,7 @@ public static void setTestingFileReader(String fileName, String type) {
                 // The layer array must be initialized before this always
             }
         } catch (FileNotFoundException ex) {
+           ex.printStackTrace();
         }
     }
 
@@ -64,6 +68,7 @@ public static void setTrainingFileReader(String fileName, String type) {
                 parameter.trainingFileReader = new MNISTDataBaseFileReader(fileName, layerArray[layerArray.length - 1]);
             }
         } catch (FileNotFoundException ex) {
+            ex.printStackTrace();
         }
     }
 
@@ -97,6 +102,8 @@ public static double rectify(double numToBeRectified) {
     }
 
     public static void setLayerArray(int... LayerArray) {
+        parameter.numberOfInputNeurons = LayerArray[0];
+        parameter.numberOfOutputNeurons = LayerArray[LayerArray.length-1];
         parameter.layerArray = LayerArray;
     }
 

output.txt

@@ -1,95 +1,16 @@
-Training accuracy in epoch 0 is 10.0
-Training accuracy in epoch 1 is 10.0
-Training accuracy in epoch 2 is 12.0
-Training accuracy in epoch 3 is 10.0
-Training accuracy in epoch 4 is 10.0
-Training accuracy in epoch 5 is 10.0
-Training accuracy in epoch 6 is 10.0
-Training accuracy in epoch 7 is 12.0
-Training accuracy in epoch 8 is 12.0
-Training accuracy in epoch 9 is 12.0
-Testing accuracy 10.13013013013013
-Label:0.0
-True Positive :0.0
-True Negative :9011.0
-False Positive :0.0
-False Negative :979.0
-
-
-Label:1.0
-True Positive :0.0
-True Negative :8856.0
-False Positive :0.0
-False Negative :1134.0
-
-
-Label:2.0
-True Positive :0.0
-True Negative :8959.0
-False Positive :0.0
-False Negative :1031.0
-
-
-Label:3.0
-True Positive :5.0
-True Negative :8896.0
-False Positive :85.0
-False Negative :1004.0
-
-
-Label:4.0
-True Positive :0.0
-True Negative :9009.0
-False Positive :0.0
-False Negative :981.0
-
-
-Label:5.0
-True Positive :0.0
-True Negative :9099.0
-False Positive :0.0
-False Negative :891.0
-
-
-Label:6.0
-True Positive :0.0
-True Negative :9033.0
-False Positive :0.0
-False Negative :957.0
-
-
-Label:7.0
-True Positive :0.0
-True Negative :8963.0
-False Positive :0.0
-False Negative :1027.0
-
-
-Label:8.0
-True Positive :0.0
-True Negative :9017.0
-False Positive :0.0
-False Negative :973.0
-
-
-Label:9.0
-True Positive :1007.0
-True Negative :89.0
-False Positive :8893.0
-False Negative :1.0
-
-
+Training accuracy in epoch 0 is 10.66891891891892
+Testing accuracy 28.92892892892893
 The neuron 31 in layer 1 has been updated by forward propagation
-Neuron #31 has activation 0.44070545272313627
+Neuron #31 has activation 6.567825572210979E-4
 
 The neuron 31 in layer 1 has been updated by forward propagation
-Neuron #31 has activation 0.805525963558943
+Neuron #31 has activation 0.003181628304117291
 
-Testing accuracy 10.128102481985588
+Testing accuracy 28.93314651721377
 The neuron 0 in layer 2 has been updated by forward propagation
-Neuron #0 has activation 0.7362380271929501
+Neuron #0 has activation 0.25373727956231534
 
 The neuron 0 in layer 2 has been updated by forward propagation
-Neuron #0 has activation 0.6814481765071231
+Neuron #0 has activation 0.7220061457959416
 
-Testing accuracy 10.136081648989393
+Testing accuracy 28.94736842105263