SVM 自定义kernel的一个简单的code 例子

function [ result_pre, result_r, max_result ] = kernel_SVM( train_data,test_data,train_label, test_label)

%train_data : m*m train-train kernel matrix

% test_data n*m test-train kernel matrix

% train_label m*1 train label

% test_label n*1 test label

r=1e-10;

for i=1:50

train_data_exp =exp(-r*train_data);

test_data_exp =exp(-r*test_data);

train_data_exp = [(1:size(index_train,2))', train_data_exp];

test_data_exp = [(1:size(index_test,2))', test_data_exp];

model_precomputed = svmtrain(train_label, train_data_exp, '-s 1 -t 4 -n 0.02 -c 1.21 -g 1');

[predict_label_P, accuracy_P, dec_values_P] = svmpredict(test_label, test_data_exp, model_precomputed);

% if (accuracy_P(1,1)>max(result_pre))

result_pre(result_index)=accuracy_P(1,1);

result_r(result_index)=r;

result_index=result_index+1;

% end

r=r*sqrt(10);

end

max_result = max(result_pre);

end