人工神经网络框架AForge学习(一)

本示例演示了单层人工神经网络的学习过程.

示例中采用了16条样本数据，3个分类，在经过7-20次的迭代后，误差值与期望最小误差(示例中为0)的差距变为最小.

  1 using  System;
  2 using  System.Collections.Generic;
  3 using  System.Text;
  4 using  System.IO;
  5
  6 using  AForge;
  7 using  AForge.Neuro;
  8 using  AForge.Neuro.Learning;
  9 using  AForge.Controls;
 10
 11 namespace  AspxOn.AForgeNeuroTest
 12 {
 13    class Program
 14    {
 15        static void Main(string[] args)
 16        {
 17            double[,] data = null;
 18            int[] classes = null;
 19
 20            double[,] tempData = new double[200, 2];
 21            int[] tempClass = new int[200];
 22
 23            double learnintRate = 0.1;
 24
 25            //最大和最小的X值
 26            double minX = double.MaxValue;
 27            double maxX = double.MinValue;
 28
 29            //样本数目统计
 30            int samples = 0;
 31            //分类数目统计
 32            int classCount = 0;
 33            int[] samplesPerClass = new int[10];
 34
 35            try
 36            {
 37                //准备训练样本数据
 38                string s = "0.1,0.1,0;0.2,0.3,0;0.3,0.4,0;0.1,0.3,0;0.2,0.5,0;0.1,1,1;0.2,1.1,1;0.3,0.9,1;";
 39                s+="0.4,0.8,1;0.2,0.9,1;1,0.4,2;0.9,0.5,2;0.8,0.6,2;0.9,0.4,2;1,0.5,2;1,0.3,2";
 40                string[] ss = s.Split(';');
 41
 42                for (int i = 0; i < ss.Length; i++)
 43                {
 44                    string str = ss[i];
 45                    string[] strs = str.Split(',');
 46
 47                    //检查数组长度
 48                    if (strs.Length != 3)
 49                        throw new Exception("length error!");
 50
 51                    //生成单词
 52                    tempData[samples, 0] = double.Parse(strs[0]);
 53                    tempData[samples, 1] = double.Parse(strs[1]);
 54                    //生成分类
 55                    tempClass[samples] = int.Parse(strs[2]);
 56
 57                    //跳过分类如果分类值大于等于10
 58                    if (tempClass[samples] >= 10)
 59                    {
 60                        continue;
 61                    }
 62
 63                    //计算不同分类的总数
 64                    if (tempClass[samples] >= classCount)
 65                        classCount += tempClass[samples] + 1;
 66
 67                    //统计每个分类的样本数目
 68                    samplesPerClass[tempClass[samples]]++;
 69
 70                    //寻找最小值
 71                    if (tempData[samples, 0] < minX)
 72                        minX = tempData[samples, 0];
 73                    //寻找最大值
 74                    if (tempData[samples, 0] > maxX)
 75                        maxX = tempData[samples, 0];
 76
 77                    //样本数目累加
 78                    samples++;
 79
 80                }
 81
 82                data = new double[samples, 2];
 83                Array.Copy(tempData, 0, data, 0, samples * 2);
 84                classes = new int[samples];
 85                Array.Copy(tempClass, 0, classes, 0, samples);
 86
 87
 88            }
 89            catch
 90            {
 91                throw new Exception("failed loading data!");
 92            }
 93
 94            /**//*=============================开始训练=================================*/
 95
 96            //准备学习样本数据
 97            double[][] input = new double[samples][];
 98            double[][] output = new double[samples][];
 99
100            for (int i = 0; i < samples; i++)
101            {
102                input[i] = new double[2];
103                output[i] = new double[classCount];
104
105                //设置输入值
106                input[i][0] = data[i, 0];
107                input[i][1] = data[i, 1];
108                //设置输出值
109                output[i][classes[i]] = 1;
110
111            }
112
113            //神经网络初始化，创建激活神经网络
114            ActivationNetwork network = new ActivationNetwork(new ThresholdFunction(), 2, classCount);
115            ActivationLayer layer = network[0];
116            //创建导师
117            PerceptronLearning teacher = new PerceptronLearning(network);
118            //设置学习比率
119            teacher.LearningRate = learnintRate;
120
121            //迭代次数
122            int iteration = 1;
123            //停止迭代标志
124            bool neetToStop = false;
125            //是否输出每次权重
126            bool showWeight = true;
127
128            try
129            {
130                //误差值
131                List<double> errors = new List<double>();
132
133                while (!neetToStop)
134                {
135                    输出权重#region 输出权重
136                    if (showWeight)
137                    {
138                        for (int i = 0; i < classCount; i++)
139                        {
140                            Console.WriteLine("neuron" + i + ";");
141                            Console.WriteLine(layer[i][0] + ";");
142                            Console.WriteLine(layer[i][1] + ";");
143                            Console.WriteLine(layer[i].Threshold.ToString());
144                            Console.WriteLine("================================");
145                        }
146                    }
147                    #endregion
148
149                    //将本次误差添加到误差列表
150                    double error = teacher.RunEpoch(input, output);
151                    errors.Add(error);
152                    //输出本次训练误差
153                    Console.WriteLine("误差:" + error.ToString());
154                    //输出当前迭代次数
155                    Console.WriteLine("迭代次数:" + iteration);
156                    Console.WriteLine("---------------------------------------------");
157
158                    //如果误差等于0,停止训练
159                    //此处可预设一个最小值，不一定非是0.当误差小于此最小值时，结束训练。
160                    if (error == 0)
161                        break;
162
163
164                    iteration++;
165                }
166            }
167            catch
168            {
169                throw new Exception("error!");
170            }
171
172            /**//*=============================结束训练=================================*/
173            
174            Console.ReadKey();
175        }
176
177    }
178}

 

输出结果如下：

转载于:https://www.cnblogs.com/waemz/archive/2009/03/02/1401350.html