Java ActivationFunction类代码示例

import org.encog.engine.network.activation.ActivationFunction; //导入依赖的package包/类
/**
 * Process one level.
 * 
 * @param currentLevel
 *            The level.
 */
private void processLevel(final int currentLevel) {
    final int fromLayerIndex = this.layerIndex[currentLevel + 1];
    final int toLayerIndex = this.layerIndex[currentLevel];
    final int fromLayerSize = this.layerCounts[currentLevel + 1];
    final int toLayerSize = this.layerFeedCounts[currentLevel];

    final int index = this.weightIndex[currentLevel];
    final ActivationFunction activation = this.getNetwork().getActivationFunctions()[currentLevel + 1];
    final double currentFlatSpot = this.flatSpot[currentLevel + 1];

    // handle weights
    int yi = fromLayerIndex;
    for(int y = 0; y < fromLayerSize; y++) {
        final double output = this.layerOutput[yi];
        double sum = 0;
        int xi = toLayerIndex;
        int wi = index + y;
        for(int x = 0; x < toLayerSize; x++) {
            this.gradients[wi] += output * this.getLayerDelta()[xi];
            sum += this.weights[wi] * this.getLayerDelta()[xi];
            wi += fromLayerSize;
            xi++;
        }

        this.getLayerDelta()[yi] = sum
                * (activation.derivativeFunction(this.layerSums[yi], this.layerOutput[yi]) + currentFlatSpot);
        yi++;
    }
}
 

import org.encog.engine.network.activation.ActivationFunction; //导入依赖的package包/类
public MLParams(double[] weights, ActivationFunction func, int nodes){
    this.weights = weights;
    this.func    = func;
    this.nodes   = nodes;
}
 

import org.encog.engine.network.activation.ActivationFunction; //导入依赖的package包/类
public void setFunc(ActivationFunction func){
    this.func = func;
}
 

import org.encog.engine.network.activation.ActivationFunction; //导入依赖的package包/类
public void addLayer(ActivationFunction func, int nodes){
    
    if(params.size() > 0) {
        buildNetwork();
    
        transformData(); 
    } else {
        intermediateDataset = this.dataset;
    }
    
    network = new BasicNetwork();
    
    network.addLayer(new BasicLayer(new ActivationLinear(), true, intermediateDataset.getInputSize()));
    
    network.addLayer(new BasicLayer(func, true, nodes));
    
    network.addLayer(new BasicLayer(new ActivationTANH(), false, intermediateDataset.getIdealSize()));
    
    network.getStructure().finalizeStructure();
    
    network.reset();
    
    train(nodes);
}
 

import org.encog.engine.network.activation.ActivationFunction; //导入依赖的package包/类
public void train(int nodes) {
    Propagation propagation = new QuickPropagation(network, intermediateDataset, 0.01);
    
    propagation.setThreadCount(0);
    
    //for(int i = 0 ; i < 100; i ++) {
        propagation.iteration();
        
        System.out.println( "In deep layer: " + params.size() + " Training error " + propagation.getError());
    //}
    
    
    int fromNodes = network.getInputCount() + 1;
    int toNodes = network.getLayerNeuronCount(1); //the next layer
    
    int numWeight = fromNodes * toNodes;
    
    double[] weights = new double[numWeight];
    int k = 0;
    for(int i = 0 ; i < fromNodes; i ++ ){
        for(int j = 0 ; j < toNodes; j ++) {
            //FIXME, bug
            weights[k++] = network.getWeight(0, i, j);
        }
    }
    
    ActivationFunction func = network.getActivation(1);
    
    
    MLParams param = new MLParams(weights, func, nodes);
    
    params.add(param);
    
    System.out.println("Add weight: " + weights.length + "\n and the activation function: " + func.toString());
}