Encog C#RBF网络,如何启动？

Question

我经历了整个文档,并没有找到如何设置RBF网络.我在ConsoleExmpales/Examples/Radial中找到了一些RBF示例,但看起来它不再起作用,因为在Encog中已经更改了一些方法.

到目前为止,我坚持这个:

    public static double[][] XORInput = {
        new[] {0.0, 0.0},
        new[] {1.0, 0.0},
        new[] {0.0, 1.0},
        new[] {1.0, 1.0}
    };

    public static double[][] XORIdeal = {
        new[] {0.0},
        new[] {1.0},
        new[] {1.0},
        new[] {0.0}
    };

        int dimension = 8;
        int numNeuronsPerDimension = 64;
        double volumeNeuronWidth = 2.0 / numNeuronsPerDimension;
        bool includeEdgeRBFs = true;

        RBFNetwork n = new RBFNetwork(dimension, numNeuronsPerDimension, 1, RBFEnum.Gaussian);
        n.SetRBFCentersAndWidthsEqualSpacing(0, 1, RBFEnum.Gaussian, volumeNeuronWidth, includeEdgeRBFs);
        //n.RandomizeRBFCentersAndWidths(0, 1, RBFEnum.Gaussian);

        INeuralDataSet trainingSet = new BasicNeuralDataSet(XORInput, XORIdeal);
        SVDTraining train = new SVDTraining(n, trainingSet);

        int epoch = 1;
        do
        {
            train.Iteration();
            Console.WriteLine("Epoch #" + epoch + " Error:" + train.Error);
            epoch++;
        } while ((epoch < 1) && (train.Error > 0.001));

当我运行这个时,我得到错误"RBF神经元的总数必须是'维度'的幂的某个整数." 在SetRBFCentersAndWidthsEqualSpacing上.如果我为RandomizeRBFCentersAndWidths更改此方法,直到达到train.iteration(),我得到"索引超出数组的范围",它可以工作.

我理解RBF网络是如何工作的,但我对SetRBFCentersAndWidthsEqualSpacing方法中的所有参数感到困惑,有人可以解释一下它的更多细节吗？

Answer 1

非常好的问题。

1. SetRBFCentersAndWidthsEqualSpacing 这是训练神经网络的相对较新的方法，Jeff Heaton 决定实施它。

2. 看起来Java 版本和C# 版本之间在第 230 - 240 行存在差异，恕我直言，错误位于 Java 版本中。

3. 我修改了您的代码，以便它可以通过附加注释运行：

   using System;
   using System.Collections.Generic;
   using System.Linq;
   using System.Text;
   using System.Threading.Tasks;
   using Encog.MathUtil.RBF;
   using Encog.Neural.Data.Basic;
   using Encog.Neural.NeuralData;
   using Encog.Neural.Rbf.Training;
   using Encog.Neural.RBF;
   
   namespace TestRBF
   {
       class Program
       {
           public static double[][] XORInput = {
           new[] {0.0, 0.0},
           new[] {1.0, 0.0},
           new[] {0.0, 1.0},
           new[] {1.0, 1.0}
       };
   
           public static double[][] XORIdeal = {
           new[] {0.0},
           new[] {1.0},
           new[] {1.0},
           new[] {0.0}
       };
   
           static void Main(string[] args)
           {
               int dimension = 2; // XORInput provides two-dimensional inputs. Not 8. 
               /*
               If XORInput is  8 dimensional  it should be like this:
   
               public static double[][] XORInput = {
               new[] {0.0, 0.0,0.0, 0.0,0.0, 0.0,0.0, 0.0}, 
               .
               .   
               .*/
               int numNeuronsPerDimension = 4; // could be also 16, 64, 256. I suppose it should accept 8, 32 but it needs additional investigation
               double volumeNeuronWidth = 2.0 / numNeuronsPerDimension;
               bool includeEdgeRBFs = true;
   
               RBFNetwork n = new RBFNetwork(dimension, numNeuronsPerDimension, 1, RBFEnum.Gaussian);
               n.SetRBFCentersAndWidthsEqualSpacing(0, 1, RBFEnum.Gaussian, volumeNeuronWidth, includeEdgeRBFs);
               //n.RandomizeRBFCentersAndWidths(0, 1, RBFEnum.Gaussian);
   
               INeuralDataSet trainingSet = new BasicNeuralDataSet(XORInput, XORIdeal);
               SVDTraining train = new SVDTraining(n, trainingSet);
   
               int epoch = 1;
               do
               {
                   train.Iteration();
                   Console.WriteLine("Epoch #" + epoch + " Error:" + train.Error);
                   epoch++;
               } while ((epoch < 1) && (train.Error > 0.001));
   
           }
       }
   }