KNN 在手写识别中的应用(Java 实现)

这篇博文主要介绍了一种基于机器学习的分类方法,K-邻近(KNN),并且使用这种方法来完成了一个简单的手写数字识别系统。

#KNN 概述

##什么是 KNN

KNN(K–nearest-neighbor),即 K-邻近算法, 所谓 K 邻近,就是 K 个最近邻居的意思,说的是每个样本都可以用与它最接近的K 个邻居来表示。

##工作原理

存在一个样本数据集合,也称作训练样本集,并且样本集中每个数据都存在标签,即我们知道样本集中每一数据与所述分类的对应关系。输入没有标签的新数据后,将新数据的每个特征与样本集中数据对应的特征进行比较,然后算法提取样本集中特征最相似数据的分类标签,一般来讲,我们只取样本集数据中前 K 个最相似的数据,最后在这 K 个数据中统计处出现次数最多的分类,最为新数据的分类。

##算法特点

  • 优点:精度高、对异常值不敏感、无数据输入假定

  • 缺点:计算复杂度高、空间复杂度高

  • 适用数据范围:数值型和标称型

##算法流程

对未知类别属性的数据集中的每个店依次执行以下操作:

  • 计算已知类别数据集中的点与当前点之间的距离

  • 按照距离递增次序排列

  • 选取与当前点距离最小的 K 个点

  • 确定前 K 个点所在类别的出现频率

  • 返回前 K 个点出现频率最高的类别作为当前点的预测分类

对于距离的计算,我们采用欧氏距离公式:

image

#KNN的应用实例 - 手写识别(Java)

##简述

我们所做的手写识别是来识别简单的手写数字,数据形式是如下图的文本文件:

image

我们有一些样本数据,然后用一些测试数据来进行算法的测试。

对于算法源码以及数据样本,详情见:https://github.com/luoyhang003/machine-learning-in-java/tree/master/k-Nearest-Neighbour

##具体实现

代码写的比较烂,只是实现了 KNN 的算法,并没有优化,敬请见谅!

  • 首先我们需要将这些文本转换为向量,可以存储于数组中
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
public static int[] data2Vec(String fileName){
int arr[] = new int[32 * 32];

try{
FileReader reader = new FileReader(fileName);
BufferedReader buffer = new BufferedReader(reader);

for(int index = 0; index < 32; index++){
String str = buffer.readLine();
int length = str.length();
for(int i = 0; i < length; i++){
String c = str.substring(i, i + 1);
arr[32*index+i] = Integer.parseInt(c);
}
}

}catch (FileNotFoundException e){
e.printStackTrace();
}catch (IOException e){
e.printStackTrace();
}

return arr;
}
  • 需要定义一个算法来计算每两个向量之间的距离
1
2
3
4
5
6
7
8
9
10
public static double calDistance(int[] a, int[] b){
double result = 0.0;
int temp = 0;
for(int i = 0; i < a.length; i++){
temp += (a[i] - b[i])*(a[i] - b[i]);
}
result = Math.sqrt(temp);

return result;
}
  • 然后我们就可以开始进行分类
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
public static int[] classify(String fileName){
int result[] = new int[2];

int arr[] = data2Vec("samples/testDigits/"+fileName);

result[0] = Integer.parseInt(fileName.split("_")[0]);

double dis[] = new double[K];
int num[] = new int[K];

for(int index = 0; index < K; index++){
dis[index] = 32;
num[index] = -1;
}

for(int i = 0; i <= 9; i++){
for(int j = 0; j < 100; j++){
int temp_arr[] = data2Vec("samples/trainingDigits/"+i+"_"+j+".txt");
double temp_dis = calDistance(arr, temp_arr);

for(int k = 0; k < K; k++){
if(temp_dis < dis[k]){
dis[k] = temp_dis;
num[k] = i;
break;
}
}
}
}

int count[] = new int[10];

for(int i = 0; i < 10; i++)
count[i] = 0;

for(int i = 0; i < K; i++){
if(num[i]!=-1)
count[num[i]]++;
}

int max = 0;
for(int i = 0; i < 10; i++){
if(count[i]>max){
max = count[i];
result[1] = i;
}
}

return result;
}
  • 最后我们来测试一下算法
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
   public static void main(String args[]){

double right = 0;
double sum = 0;

for(int i = 0; i < 10; i++){
for(int j = 0; j < 50; j++){
int result[] = classify(""+i+"_"+j+".txt");

System.out.println("the classifier came back with: "+result[1]+" , the real answer is: " +result[0]);
sum++;
if(result[0]==result[1])
right++;
}
}
System.out.println("right:"+right);
System.out.println("sum:"+sum);

double rate = right/sum;
System.out.println("the total right rate is: " + rate);

}

得到的结果是:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
the classifier came back with: 0 , the real answer is: 0
the classifier came back with: 0 , the real answer is: 0
the classifier came back with: 0 , the real answer is: 0
the classifier came back with: 0 , the real answer is: 0
……


the classifier came back with: 9 , the real answer is: 9
the classifier came back with: 9 , the real answer is: 9
the classifier came back with: 9 , the real answer is: 9
the classifier came back with: 9 , the real answer is: 9
the classifier came back with: 9 , the real answer is: 9
right:486.0
sum:500.0
the total right rate is: 0.972

完整的代码与测试数据详见:https://github.com/luoyhang003/machine-learning-in-java/tree/master/k-Nearest-Neighbour


版权声明:本文为博主原创文章,未经博主允许不得转载。

文章来源:http://blog.luoyuanhang.com