Using Neural Network Class in WEKA in Java code

Question

Hi I want to do simple training and testing using Neural Network in WEKA library.

But, I find it is not trivial, and its different with NaiveBayes class in its library.

Anyone have example how to use this class in java code?

Answer 1

Following steps might be able to help you:

1. Add Weka libraries

Download Weka from http://www.cs.waikato.ac.nz/ml/weka/downloading.html.

From the package find 'Weka.jar' and add in the project.

Java Code Snippet

2. Building a Neural Classifier

   public void simpleWekaTrain(String filepath)
   {
   try{
   //Reading training arff or csv file
   FileReader trainreader = new FileReader(filepath);
   Instances train = new Instances(trainreader);
   train.setClassIndex(train.numAttributes() – 1);
   //Instance of NN
   MultilayerPerceptron mlp = new MultilayerPerceptron();
   //Setting Parameters
   mlp.setLearningRate(0.1);
   mlp.setMomentum(0.2);
   mlp.setTrainingTime(2000);
   mlp.setHiddenLayers(“3?);
   mlp.buildClassifier(train);
   }
   catch(Exception ex){
   ex.printStackTrace();
   }
   }
   

Another Way to set parameters,

    mlp.setOptions(Utils.splitOptions(“-L 0.1 -M 0.2 -N 2000 -V 0 -S 0 -E 20 -H 3?));

Where,

L = Learning Rate
M = Momentum
N = Training Time or Epochs
H = Hidden Layers
etc.

3. Neural Classifier Training Validation

For evaluation of training data,

    Evaluation eval = new Evaluation(train);
    eval.evaluateModel(mlp, train);
    System.out.println(eval.errorRate()); //Printing Training Mean root squared Error
    System.out.println(eval.toSummaryString()); //Summary of Training

To apply K-Fold validation

    eval.crossValidateModel(mlp, train, kfolds, new Random(1));

4. Evaluating/Predicting unlabelled data

   Instances datapredict = new Instances(
   new BufferedReader(
   new FileReader(<Predictdatapath>)));
   datapredict.setClassIndex(datapredict.numAttributes() – 1);
   Instances predicteddata = new Instances(datapredict);
   //Predict Part
   for (int i = 0; i < datapredict.numInstances(); i++) {
   double clsLabel = mlp.classifyInstance(datapredict.instance(i));
   predicteddata.instance(i).setClassValue(clsLabel);
   }
   //Storing again in arff
   BufferedWriter writer = new BufferedWriter(
   new FileWriter(<Output File Path>));
   writer.write(predicteddata.toString());
   writer.newLine();
   writer.flush();
   writer.close();
   

Answer 2

I read some sources on the internet and just realize that "if you want to use NeuralNetwork classifier in WEKA library, so the approach is NOT using the given NeuralNetwork class, but it should be "MultilayerPerceptron" class"

It's a little bit tricky and consumed my hours.

I hope it's useful for anyone who is struggling with this.

http://weka.8497.n7.nabble.com/Multi-layer-perception-td2896.html

Ps. Please correct if I am being wrong!