NeuroPh Error won't diminish - artificial-intelligence

Question

I am creating a network for water level forecasting. I am using NeuroPh 2.91 for windows. I set the network to 3 inputs since it is accepting 3 inputs namely water level, rainfall, and inflow. I am using multi-layer perceptron, tanh as transfer function and backpropagation as learning rule with 9 hidden neurons.
I am always having this output:
Starting neural network training...
Training network try using data set adminshet
Training error: null
and the total network error according to the graph is 20,000+ .
What should i do? I am really new to ANN and Neuroph.

Answer 1

I've had the same issue here. Similar setup. It works for me, whe I strongly limit the Max-Iterations. e.g. to 10.
This makes me think, that there is a bug in the NeurophStudio.
Short Tip, which worked for me:
Do it yourself! Open Eclipse, add a Project, add the Neuroph jars and build your network. Its hard, but this works exactly as expected. You have to dump your own results into a csv file and display it with Excel. But ANN-handling doesn't work just by "clicking on a gui".
package de.sauer.dispe;
import org.neuroph.core.Layer;
import org.neuroph.core.NeuralNetwork;
import org.neuroph.core.Neuron;
import org.neuroph.core.data.DataSet;
import org.neuroph.core.transfer.Linear;
import org.neuroph.core.transfer.Tanh;
import org.neuroph.nnet.MultiLayerPerceptron;
import org.neuroph.nnet.learning.BackPropagation;
public class DirSpeCntrl {
private static final int MAX_ITER = 2000;
private static final double MAX_ERROR = 0.005;
private static final double LEARNING_RATE = 0.1;
public static void main(String[] args) {
System.out.println("Create ANN");
NeuralNetwork<BackPropagation> nn = new MultiLayerPerceptron(3, 15, 15, 1);
// Setting ALL neurons to TanH transferfunction (important, if you have negativ values)
Layer[] layers = nn.getLayers();
for(Layer curLayer: layers) {
for(Neuron curNeuron: curLayer.getNeurons()) {
curNeuron.setTransferFunction(new Tanh());
}
}
for(Neuron curNeuron: layers[3].getNeurons()) {
curNeuron.setTransferFunction(new Linear());
}
nn.randomizeWeights();
System.out.println("Load Sampledata...");
DataSet ds = DataSet.createFromFile(
"C:\\Users\\meist_000\\Documents\\Thesis\\vanilla_eng.csv",
3, 1, ";");
System.out.println("done: "+ds.getRows().size()+". Learn...");
// Setting stuff
BackPropagation lr = new BackPropagation();
lr.setLearningRate(LEARNING_RATE);
lr.setMaxIterations(MAX_ITER);
lr.setTrainingSet(ds);
lr.setNeuralNetwork(nn);
nn.setLearningRule(lr);
// bla.learn(ds); Faster bulk operation...
// Slower single operation with logging:
for(int i=0;i<MAX_ITER;i++) {
lr.doLearningEpoch(ds);
double curError = lr.getTotalNetworkError();
System.out.println(curError);
if(curError < MAX_ERROR) {
System.out.println("Stopped on "+i);
break;
}
}
// Testing the network
nn.setInput(new double[] {0.080484492, -0.138512128, -0.140826873});
nn.calculate();
double[] prediction = nn.getOutput();
System.out.println("Pred: "+prediction[0]);
}
}