SVM 分类器

6091820503

最近在学习Machine learning，尝试自己用Matlab实现SVM算法
svm基本思路是让分界的margin尽量大，可以转化为一个standard quadratic programming problem,这样在matlab中有就可以调用现成的函数解决


Code

1. %svm demo
   
2. %this m file can be used to solve unseparable problem
   
3. clear;
   
4. close all;
   
5. clc;
   
6. 
   
7. %positive points numbers
   
8. Np = 15 ;
   
9. %negative points numbers
   
10. Nn = 15 ;
    
11. %generate data
    
12. [x, y] = gendata(Np, Nn) ;
    
13. 
    
14. tol = 1e-4 ;
    
15. 
    
16. 
    
17. %% build matrix for quadprog()
    
18. [n m] = size(x); %m: number of train data; n: length of xi
    
19. Y = diag(y);
    
20. %%primal formulation
    
21. %linear kernel
    
22. K = x'*x;
    
23. C = 1;
    
24. H = Y*K*Y;
    
25. f = -ones(m, 1);
    
26. A = [];
    
27. b = [];
    
28. 
    
29. Aeq = y;
    
30. beq = 0;
    
31. 
    
32. l = zeros(m, 1);
    
33. u = 1*ones(m, 1);
    
34. options = optimset('Algorithm','interior-point-convex');
    
35. 
    
36. %%standard quadratic programming problem X = quadprog(H,f,A,b,Aeq,beq,LB,UB,X0)
    
37. alpha = quadprog(H, f, A, b, Aeq, beq, l, u, [], options);
    
38. 
    
39. %%compute w
    
40. w = zeros(n, 1);
    
41. for i = 1 : m
    
42.         w = w + alpha(i)*y(i)*x(:, i);
    
43. end
    
44. %%compute b
    
45. pos = find(alpha > 1e-6);
    
46. b = mean(y(pos)- w'*x(:, pos));
    
47. 
    
48. 
    
49. %%plot the boundry on the map
    
50. xt = linspace(min(x(1, :)), max(x(2, :)), 100);
    
51. yt = (-b-w(1)*xt)/w(2);
    
52. yt_p = (1 -b -w(1)*xt)/w(2);
    
53. yt_n = (-1 -b - w(1)*xt)/w(2);
    
54. 
    
55. %%plot data
    
56. xp = x(:, y == 1);
    
57. xn = x(:, y == -1);
    
58. figure;
    
59. plot(xp(1, :), xp(2, :), 'ob', xn(1, :), xn(2, :), '*r');
    
60. hold on;
    
61. plot(xt, yt, '-r', xt, yt_n, '-b', xt, yt_p, '-b');
    
62. hold off;
    
63. print('-dpng', 'svm_soft_margin_simu.png')
    

复制代码

6091820503

占个楼先        

6091820503

自己占个楼先

RUANJI

MATLAB已经有现成的SVM分类器可以用了。

LIBSVM是也不错，国内用的较多。

6091820503

RUANJI 发表于 2015-1-9 22:20
MATLAB已经有现成的SVM分类器可以用了。

LIBSVM是也不错，国内用的较多。

还是感觉用自己写的库最靠谱