有没有会支持向量机的，本人有代码讲解改错，价钱可商量

Question

本人有一个支持向量机代码的包，本人可讲解该包的理论需要一个会编写matlab程序的帮我讲解一下并处理一下错误
下面是代码，价钱可商量

Answer 1

function [label,prediction,cpu_time] = means3vm_iter(y,x,x_test,opt)

% means3vm_iter implements the means3vm_iter algorithm as shown in [1].
%
%: means3vm_iter employs the Matlab version of libsvm [2] (available at
% http://www.csie.ntu.edu.tw/~cjlin/libsvm/) and a slight modified matlab
% version of libsvm (availaible at 'libsvm-mat-2.83-1-[modified]' included).
%
% Syntax
%
% [label,prediction,cputime] = means3vm_iter(y,x,x_test,opt)
%
% Description
%
% means3vm_iter takes,
% y - A nx1 label vector, where y = 1 means positive, y = -1 means negative, y = 0 means unlabeled
% x - A nxd training data matrix, where d is the dimension of instance
% x_test - A mxd testing data matrix
% opt - A structure describes option of SVM
% 1) opt.c1: regularization term for labeled instances (see Eq.(7) in [1]), default setting opt.c1= 100
% 2) opt.c2: regularization term for unlabeled instances(see Eq.(7) in [1]), default setting opt.c2 = 0.1
% 3) opt.gaussian: kernel type, 1 means gaussian kernel, 0 means linear kernel, default setting opt.gaussian = 0
% 4) opt.gamma: parameter for gaussian kernel, i.e.,k(x,y) = exp(-gamma*||x-y||^2), default settting 1/gamma = average distance between pattern
% 5) opt.maxiter: maximal iteration number, default setting opt.maxiter = 50
% 6) opt.ep: expected number of positive instance among unlabeled data, default setting opt.ep = prior from labeled data
% and returns,
% label - A mx1 label vector, the predicted label of the testing data
% prediction - A mx1 prediction vector, the prediction of the testing data
% cputime - cpu running time
%
%
% [1] Y.-F. Li, J. T. Kwok, and Z.-H. Zhou. Semi-supervised learning using label mean. In: Proceedings of the 26th International Conference on Machine Learning (ICML'09), Montreal, Canada, 2009, pp.633-640.
% [2] R.-E. Fan, P.-H. Chen, and C.-J. Lin. Working set selection using second order information for training SVM. Journal of Machine Learning Research 6, 1889-1918, 2005.

tt = cputime;
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%1. initilization
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

% 1.1 initilize the option
n = size(x,1);
m = size(x_test,1);
l_inx = find(y);
u_inx = find(y==0);
ln = length(l_inx);
un = length(u_inx);

if ~isfield(opt,'c1')
opt.c1 = 100;
end
if ~isfield(opt,'c2')
opt.c2 = 0.1;
end
if ~isfield(opt,'gaussian')
opt.gaussian = 0;
end
if opt.gaussian == 1 && ~isfield(opt,'gamma')
opt.gamma = n^2/sum(sum(repmat(sum(x.x,2)',n,1) + repmat(sum(x.x,2),1,n) ...
- 2xx'));
end
if ~isfield(opt,'maxiter')
opt.maxiter = 50;
end
if ~isfield(opt,'ep')
opt.ep = ceil(length(find(y == 1))/ln*un);
end

%1.2 calculate the kernel matrix
if opt.gaussian == 1
K = exp(- (repmat(sum(x.x,2)',n,1) + repmat(sum(x.x,2),1,n)-2xx') opt.gamma);
K = K + 1e-10eye(size(K,1));
K_test = exp(- (repmat(sum(x.x,2),1,m) + repmat(sum(x_test.x_test,2)',n,1)- 2xx_test') * opt.gamma);
else
K = xx';
K = K + 1e-10eye(size(K,1));
K_test = x*x_test';
end

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% 2. train supervised SVM
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
opt_svm = ['-t 4 -c ' num2str(opt.c1)];
K_l = K(l_inx,l_inx);
K_l = [(1:ln)',K_l];
addpath('libsvm-mat-2.83-1');
model = svmtrain(y(l_inx),K_l,opt_svm);

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% 3. iteratively predict the unlabeled data and refine svm
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

% 3.1 predict the label for unlabeled data
K_t = K(l_inx,u_inx);
K_t = [(1:un)',K_t'];
[predict_label, accuracy, dec_values] = svmpredict(ones(un,1),K_t,model);
if model.Label(1) == -1
dec_values = -dec_values;
end
tmpd = y;
[val,ix] = sort(dec_values,'descend');
tmpd(u_inx(ix(1:opt.ep))) = 1;
tmpd(u_inx(ix(opt.ep+1:un))) = -1;

iter = 1; flag = 1;
while iter < opt.maxiter && flag

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% 3.2 Solve a quadratic programming problem (Eq.(8) in [1]) which is dual form of Eq.(7) in [1]
% 
%        max_a  \sum_{i=1}^{ln} a_i - 1/2 (a.*\hat{y})' K^{tmpd} (a.* \hat{y})
%        s.t.   \sum_{i=1}^{ln} a_i + a_{ln+1} - a_{ln+2} = 0,
%               a_{ln+1} + a_{ln+2} = c2,
%               0 <= a_i <= c1, i = 1,..., ln
%               0 <= a_i <= c2, i = ln+1, ln+2
% where K^{tmpd} is defined as the same as Eq.(10) in [1] and \hat{y} = [y,1,-1];
% Here, we solve this QP problem via standard matlab function 'quadprog'
%   QUADPROG Quadratic programming. 
%   X = QUADPROG(H,f,A,b) attempts to solve the quadratic programming
%   problem:
%
%            min 0.5*x'*H*x + f'*x   subject to:  A*x <= b
%             x
%
%   X = QUADPROG(H,f,A,b,Aeq,beq) solves the problem above while
%   additionally satisfying the equality constraints Aeq*x = beq.
%
%   X = QUADPROG(H,f,A,b,Aeq,beq,LB,UB) defines a set of lower and upper
%   bounds on the design variables, X, so that the solution is in the
%   range LB <= X <= UB. Use empty matrices for LB and UB if no bounds
%   exist. Set LB(i) = -Inf if X(i) is unbounded below; set UB(i) = Inf if
%   X(i) is unbounded above.

var_n = ln+2;   % variable number 
con_n = 2;      % constraint number

pos_inx = u_inx(logical(tmpd(u_inx)==1));
neg_inx = u_inx(logical(tmpd(u_inx)==-1));

% Hessian matrix, i.e., H = K^{tmpd}.* (\hat{y}*\hat{y}')
H = zeros(var_n);
H(1:ln,1:ln) = K(l_inx,l_inx); 
H(ln+1,1:ln) = mean(K(pos_inx,l_inx));
H(1:ln,ln+1) = H(ln+1,1:ln)';
H(ln+2,1:ln) = mean(K(neg_inx,l_inx));
H(1:ln,ln+2) = H(ln+2,1:ln)';
H(ln+1,ln+1) = mean(mean(K(pos_inx,pos_inx)));
H(ln+1,ln+2) = mean(mean(K(pos_inx,neg_inx)));
H(ln+2,ln+1) = H(ln+1,ln+2);
H(ln+2,ln+2) = mean(mean(K(neg_inx,neg_inx)));      
tr_y = [y(l_inx);1;-1];
H = H.* (tr_y*tr_y');
for i = 1:var_n
    H(i,i) = H(i,i) + 1e-6;
end

%linear term f
f = zeros(var_n,1);
f(1:ln) = -1;
    
%Aeq
A = zeros(con_n, var_n);
A(1,1:var_n) = [y(l_inx)',1,-1];
A(2,ln+1:ln+2) = 1;
           
%b
b = [0;opt.c2];
    
%lb & ub
lb = zeros(var_n,1);
ub = opt.c1*ones(var_n,1);
ub(ln+1:ln+2) = opt.c2;
    
%call quadprog
tic
alpha = quadprog(H,f,[],[],A,b,lb,ub);
toc
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% 3.1 again, predict the label for unlabeled data

% get the Support Vector
inx = find(1e-6 < alpha(1:ln));  

% calculate the rho, f(x) = w'x + rho
pre_val = (alpha.*tr_y)'*(H.*(tr_y*tr_y'));
rho = mean(tr_y(inx)-pre_val(inx)');

% reconstruct the prdiction
new_alpha = zeros(1,n);
new_alpha(l_inx) = alpha(1:ln)';
new_alpha(pos_inx) = alpha(ln+1)/length(pos_inx);
new_alpha(neg_inx) = alpha(ln+2)/length(neg_inx);      
pre_val = (new_alpha.*tmpd')*K + rho;

% predict the label for unlabeled data
tmptmpd = y;
[val,ix] = sort(pre_val(u_inx),'descend');
tmptmpd(u_inx(ix(1:opt.ep))) = 1;
tmptmpd(u_inx(ix(opt.ep+1:un))) = -1;

% check the stop condition
if tmptmpd == tmpd
    flag = 0;
else
    tmpd = tmptmpd; iter = iter + 1;
end  

end

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% 4. train a final SVM and predict the test data
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% preprocess the input for means3vm
addpath('libsvm-means3vm')
posinx = find(tmpd == 1); pos = length(posinx);
neginx = find(tmpd == -1); neg = length(neginx);

pl = length(find(y(l_inx) == 1));
nl = length(find(y(l_inx) == -1));
posinx(1:pl) = find(y == 1);
posinx(pl+1:pos) = u_inx(logical(tmpd(u_inx)'==1));
neginx(1:nl) = find(y == -1);
neginx(nl+1:neg) = u_inx(logical(tmpd(u_inx)' == -1));

tmpy = tmpd;
tmpy(1:pos) = 1;
tmpy(pos+1:pos+neg) = -1;

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% train means3vm
opt_svm = ['-t 4 -c ' num2str(opt.c1) ' -w2 ' num2str(opt.c2/opt.c1)];
K_l = K([posinx;neginx],[posinx;neginx]);
K_l = [(1:n)',K_l];
model = svmtrain(tmpy,K_l,opt_svm,[pl,nl,pos,neg]);

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% prediction
K_test = [(1:m)',K_test([posinx;neginx],:)'];
yy_test = zeros(m,1);
[label, accuracy, prediction] = svmpredict(yy_test,K_test,model)
if model.Label(1) <0
prediction = -prediction;
end
cpu_time = cputime - tt;