Search code examples
javaneural-networkbackpropagation

Error NullPointer Neuroph doOneLearningIteration

I'm using Neuroph 2.9 framework to code ANN to predict housing prices. I want get every error every time run every epoch (to show the improve of error on chart) but this cause error.

// create multi layer perceptron
    System.out.println("Creating neural network");
    MultiLayerPerceptron neuralNet = new MultiLayerPerceptron(
            TransferFunctionType.SIGMOID, inputsCount, hiddentsCount1,
            outputsCount);

// set learning parameters
    MomentumBackpropagation learningRule = new MomentumBackpropagation();
    learningRule.setLearningRate(0.3);
    learningRule.setMomentum(0.5);
    learningRule.setNeuralNetwork(neuralNet);

    learningRule.setTrainingSet(TrainSet);
    learningRule.doOneLearningIteration(TrainSet);

I get this:

Exception in thread "main" java.lang.NullPointerException
at org.neuroph.nnet.learning.MomentumBackpropagation.updateNeuronWeights(MomentumBackpropagation.java:72)
at org.neuroph.nnet.learning.BackPropagation.calculateErrorAndUpdateOutputNeurons(BackPropagation.java:83)
at org.neuroph.nnet.learning.BackPropagation.updateNetworkWeights(BackPropagation.java:53)
at org.neuroph.core.learning.SupervisedLearning.learnPattern(SupervisedLearning.java:190)
at org.neuroph.core.learning.SupervisedLearning.doLearningEpoch(SupervisedLearning.java:165)
at org.neuroph.core.learning.IterativeLearning.doOneLearningIteration(IterativeLearning.java:245)
at com.thao.Main.main(Main.java:76)

The problem is when I use : learningRule.learn(TrainSet); it's ok, no error come out. The documentation so bad to differ functions to chose right function to run the right thing I want.

Solution

  • The thing I found is that doOneLearningIteration function didn't work because inside them. It's not initiated. Therefore, to run, we need to override or run 1 epoch and then doOneLearningIteration. That's work for me.