基于Java實(shí)現(xiàn)的一層簡(jiǎn)單人工神經(jīng)網(wǎng)絡(luò)算法示例

import java.util.Arrays;
public class Data {
  double[] vector;
  int dimention;
  int type;
  public double[] getVector() {
    return vector;
  }
  public void setVector(double[] vector) {
    this.vector = vector;
  }
  public int getDimention() {
    return dimention;
  }
  public void setDimention(int dimention) {
    this.dimention = dimention;
  }
  public int getType() {
    return type;
  }
  public void setType(int type) {
    this.type = type;
  }
  public Data(double[] vector, int dimention, int type) {
    super();
    this.vector = vector;
    this.dimention = dimention;
    this.type = type;
  }
  public Data() {
  }
  @Override
  public String toString() {
    return "Data [vector=" + Arrays.toString(vector) + ", dimention=" + dimention + ", type=" + type + "]";
  }
}

package cn.edu.hbut.chenjie;
import java.util.ArrayList;
import java.util.List;
import java.util.Random;
import org.jfree.chart.ChartFactory;
import org.jfree.chart.ChartFrame;
import org.jfree.chart.JFreeChart;
import org.jfree.data.xy.DefaultXYDataset;
import org.jfree.ui.RefineryUtilities;
public class ANN2 {
  private double eta;//學(xué)習(xí)率
  private int n_iter;//權(quán)重向量w[]訓(xùn)練次數(shù)
  private List<Data> exercise;//訓(xùn)練數(shù)據(jù)集
  private double w0 = 0;//閾值
  private double x0 = 1;//固定值
  private double[] weights;//權(quán)重向量，其長(zhǎng)度為訓(xùn)練數(shù)據(jù)維度+1，在本例中數(shù)據(jù)為2維，故長(zhǎng)度為3
  private int testSum = 0;//測(cè)試數(shù)據(jù)總數(shù)
  private int error = 0;//錯(cuò)誤次數(shù)
  DefaultXYDataset xydataset = new DefaultXYDataset();
  /**
   * 向圖表中增加同類型的數(shù)據(jù)
   * @param type 類型
   * @param a 所有數(shù)據(jù)的第一個(gè)分量
   * @param b 所有數(shù)據(jù)的第二個(gè)分量
   */
  public void add(String type,double[] a,double[] b)
  {
    double[][] data = new double[2][a.length];
    for(int i=0;i<a.length;i++)
    {
      data[0][i] = a[i];
      data[1][i] = b[i];
    }
    xydataset.addSeries(type, data);
  }
  /**
   * 畫圖
   */
  public void draw()
  {
    JFreeChart jfreechart = ChartFactory.createScatterPlot("exercise", "x1", "x2", xydataset);
    ChartFrame frame = new ChartFrame("訓(xùn)練數(shù)據(jù)", jfreechart);
    frame.pack();
    RefineryUtilities.centerFrameOnScreen(frame);
    frame.setVisible(true);
  }
  public static void main(String[] args)
  {
    ANN2 ann2 = new ANN2(0.001,100);//構(gòu)造人工神經(jīng)網(wǎng)絡(luò)
    List<Data> exercise = new ArrayList<Data>();//構(gòu)造訓(xùn)練集
    //人工模擬1000條訓(xùn)練數(shù)據(jù) ，分界線為x2=x1+0.5
    for(int i=0;i<1000000;i++)
    {
      Random rd = new Random();
      double x1 = rd.nextDouble();//隨機(jī)產(chǎn)生一個(gè)分量
      double x2 = rd.nextDouble();//隨機(jī)產(chǎn)生另一個(gè)分量
      double[] da = {x1,x2};//產(chǎn)生數(shù)據(jù)向量
      Data d = new Data(da, 2, x2 > x1+0.5 ? 1 : -1);//構(gòu)造數(shù)據(jù)
      exercise.add(d);//將訓(xùn)練數(shù)據(jù)加入訓(xùn)練集
    }
    int sum1 = 0;//記錄類型1的訓(xùn)練記錄數(shù)
    int sum2 = 0;//記錄類型-1的訓(xùn)練記錄數(shù)
    for(int i = 0; i < exercise.size(); i++)
    {
      if(exercise.get(i).getType()==1)
        sum1++;
      else if(exercise.get(i).getType()==-1)
        sum2++;
    }
    double[] x1 = new double[sum1];
    double[] y1 = new double[sum1];
    double[] x2 = new double[sum2];
    double[] y2 = new double[sum2];
    int index1 = 0;
    int index2 = 0;
    for(int i = 0; i < exercise.size(); i++)
    {
      if(exercise.get(i).getType()==1)
      {
        x1[index1] = exercise.get(i).vector[0];
        y1[index1++] = exercise.get(i).vector[1];
      }
      else if(exercise.get(i).getType()==-1)
      {
        x2[index2] = exercise.get(i).vector[0];
        y2[index2++] = exercise.get(i).vector[1];
      }
    }
    ann2.add("1", x1, y1);
    ann2.add("-1", x2, y2);
    ann2.draw();
    ann2.input(exercise);//將訓(xùn)練集輸入人工神經(jīng)網(wǎng)絡(luò)
    ann2.fit();//訓(xùn)練
    ann2.showWeigths();//顯示權(quán)重向量
    //人工生成一千條測(cè)試數(shù)據(jù)
    for(int i=0;i<10000;i++)
    {
      Random rd = new Random();
      double x1_ = rd.nextDouble();
      double x2_ = rd.nextDouble();
      double[] da = {x1_,x2_};
      Data test = new Data(da, 2, x2_ > x1_+0.5 ? 1 : -1);
      ann2.predict(test);//測(cè)試
    }
    System.out.println("總共測(cè)試" + ann2.testSum + "條數(shù)據(jù),有" + ann2.error + "條錯(cuò)誤,錯(cuò)誤率:" + ann2.error * 1.0 /ann2.testSum * 100 + "%");
  }
  /**
   *
   * @param eta 學(xué)習(xí)率
   * @param n_iter 權(quán)重分量學(xué)習(xí)次數(shù)
   */
  public ANN2(double eta, int n_iter) {
    this.eta = eta;
    this.n_iter = n_iter;
  }
  /**
   * 輸入訓(xùn)練集到人工神經(jīng)網(wǎng)絡(luò)
   * @param exercise
   */
  private void input(List<Data> exercise) {
    this.exercise = exercise;//保存訓(xùn)練集
    weights = new double[exercise.get(0).dimention + 1];//初始化權(quán)重向量，其長(zhǎng)度為訓(xùn)練數(shù)據(jù)維度+1
    weights[0] = w0;//權(quán)重向量第一個(gè)分量為w0
    for(int i = 1; i < weights.length; i++)
      weights[i] = 0;//其余分量初始化為0
  }
  private void fit() {
    for(int i = 0; i < n_iter; i++)//權(quán)重分量調(diào)整n_iter次
    {
      for(int j = 0; j < exercise.size(); j++)//對(duì)于訓(xùn)練集中的每條數(shù)據(jù)進(jìn)行訓(xùn)練
      {
        int real_result = exercise.get(j).type;//y
        int calculate_result = CalculateResult(exercise.get(j));//y'
        double delta0 = eta * (real_result - calculate_result);//計(jì)算閾值更新
        w0 += delta0;//閾值更新
        weights[0] = w0;//更新w[0]
        for(int k = 0; k < exercise.get(j).getDimention(); k++)//更新權(quán)重向量其它分量
        {
          double delta = eta * (real_result - calculate_result) * exercise.get(j).vector[k];
          //Δw=η*(y-y')*X
          weights[k+1] += delta;
          //w=w+Δw
        }
      }
    }
  }
  private int CalculateResult(Data data) {
    double z = w0 * x0;
    for(int i = 0; i < data.dimention; i++)
      z += data.vector[i] * weights[i+1];
    //z=w0x0+w1x1+...+WmXm
    //激活函數(shù)
    if(z>=0)
      return 1;
    else
      return -1;
  }
  private void showWeigths()
  {
    for(double w : weights)
      System.out.println(w);
  }
  private void predict(Data data) {
    int type = CalculateResult(data);
    if(type == data.getType())
    {
      //System.out.println("預(yù)測(cè)正確");
    }
    else
    {
      //System.out.println("預(yù)測(cè)錯(cuò)誤");
      error ++;
    }
    testSum ++;
  }
}

-0.22000000000000017
-0.4416843982815453
0.442444202054685
總共測(cè)試10000條數(shù)據(jù),有17條錯(cuò)誤,錯(cuò)誤率:0.16999999999999998%