标题: knn算法的MATLAB源码 [打印本页]

---

作者: lixiansheng    时间: 2018-6-9 15:57
标题: knn算法的MATLAB源码

1. %knn 归一化自写
   
2. 
   
3. % 把数据分为测试集和训练集
   
4. % train=[wine(1:30,:);wine(60:89,:);wine(131:160,:)];
   
5. % test=[wine(31:59,:);wine(90:130,:);wine(161:178,:)];
   
6. % train_labels=[wine_labels(1:30,:);wine_labels(60:89,:);wine_labels(131:160,:)];
   
7. % test_labels=[wine_labels(31:59,:);wine_labels(90:130,:);wine_labels(161:178,:)];
   
8. %把标签加到数据的最后一列
   
9. wine(:,14)=wine_labels;
   
10. %按行随机打乱数据
    
11. wine_data =wine(randperm(size(wine,1)),:);
    
12. %取前一百条数据给训练集
    
13. train=wine_data(1:100,1:13);
    
14. train_labels=wine_data(1:100,14);
    
15. %剩下的78条给测试集
    
16. test=wine_data(101:178,1:13);
    
17. test_labels=wine_data(101:178,14);
    
18. %数据归一化
    
19. h=max(train,[],1);
    
20. l=min(train,[],1);
    
21. for i=1:13
    
22.     train(:,i)=(train(:,i)-l(i))./(h(i)-l(i));
    
23.     test(:,i)=(test(:,i)-l(i))./(h(i)-l(i));
    
24. end
    
25. %[train,strc]=mapminmax(train');  
    
26. %strc归一化返回的参数 max min
    
27. %test=mapminmax.apply(test',strc);
    
28. %计算测试集与所有训练集间的距离
    
29. all_distance=dist(test,train');
    
30. %升序排序 B为返回的索引
    
31. [A,B]=sort(all_distance,2);
    
32. %Knn  K近邻
    
33. K=3;
    
34. all_labels=train_labels(B(:,1:K));
    
35. %统计矩阵中出现的类别
    
36. b=unique(all_labels);
    
37. %统计各个类别出现的次数
    
38. c=histc(all_labels',b);
    
39. %新标签赋值
    
40. [m,predict_labels]=max(c);
    
41. %求准确率
    
42. accuracy=1-(length(find((predict_labels-test_labels')~=0))/length(test_labels))
    
43. %画图
    
44. figure;
    
45. hold on;
    
46. plot(test_labels,'o');
    
47. plot(predict_labels,'*');
    
48. xlabel('测试集数据','FontSize',12);
    
49. ylabel('类别标签','FontSize',12);
    
50. legend('实际测试集分类','预测测试集分类');
    
51. title('测试集的实际分类和预测分类图','FontSize',12);
    
52. grid on;
    
53. 
    
54. 
    
55. 
    
56. 
    
57. 
    
58. %knn 归一化使用mapminmax函数进行的
    
59. %knn
    
60. 
    
61. % 把数据分为测试集和训练集
    
62. % train=[wine(1:30,:);wine(60:89,:);wine(131:160,:)];
    
63. % test=[wine(31:59,:);wine(90:130,:);wine(161:178,:)];
    
64. % train_labels=[wine_labels(1:30,:);wine_labels(60:89,:);wine_labels(131:160,:)];
    
65. % test_labels=[wine_labels(31:59,:);wine_labels(90:130,:);wine_labels(161:178,:)];
    
66. %把标签加到数据的最后一列
    
67. wine(:,14)=wine_labels;
    
68. %按行随机打乱数据
    
69. wine_data =wine(randperm(size(wine,1)),:);
    
70. %取前一百条数据给训练集
    
71. train=wine_data(1:100,1:13);
    
72. train_labels=wine_data(1:100,14);
    
73. %剩下的78条给测试集
    
74. test=wine_data(101:178,1:13);
    
75. test_labels=wine_data(101:178,14);
    
76. %数据归一化
    
77. [train,strc]=mapminmax(train');  
    
78. %strc归一化返回的参数 max min
    
79. test=mapminmax.apply(test',strc);
    
80. %计算测试集与所有训练集间的距离
    
81. all_distance=dist(test',train);
    
82. %升序排序 B为返回的索引
    
83. [A,B]=sort(all_distance,2);
    
84. %Knn  K近邻
    
85. K=5;
    
86. all_labels=train_labels(B(:,1:K));
    
87. %统计矩阵中出现的类别
    
88. b=unique(all_labels);
    
89. %统计各个类别出现的次数
    
90. c=histc(all_labels',b);
    
91. %新标签赋值
    
92. [~,predict_labels]=max(c);
    
93. %求准确率
    
94. accuracy=1-(length(find((predict_labels-test_labels')~=0))/length(test_labels))
    
95. %画图
    
96. figure;
    
97. hold on;
    
98. plot(test_labels,'o');
    
99. plot(predict_labels,'*');
    
100. xlabel('测试集数据','FontSize',12);
     
101. ylabel('类别标签','FontSize',12);
     
102. legend('实际测试集分类','预测测试集分类');
     
103. title('测试集的实际分类和预测分类图','FontSize',12);
     
104. grid on;
     

复制代码

---

欢迎光临 (http://www.51hei.com/bbs/)

Powered by Discuz! X3.1