%?my_seqminopt
????%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
????%請使用自己設計的SMO算法替代原Matlab的SMO算法
%?????[alpha?bias]?=?seqminopt（training?groupIndex?...
%?????????boxconstraint?tmp_kfun?smo_opts）;
????%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
????
function?[alpha?bias]?=?my_seqminopt（training_data?Labels?boxconstraint?kernel_func?smo_opts）

N?=?length（training_data）;%計算樣本數量
alpha?=?zeros（N1）;%初始話所有的拉格朗日橙子
bias?=?0.0;%初始化偏移量

num_changed?=?0;%拉格朗日橙子改變的個數
exam_all?=?1;%是否檢查全部乘子

while?num_changed?>0?||?exam_all
?????num_changed?=?0;
?????if?exam_all==1
??????????for?i=1:N?
???????????????num_changed?=?num_changed?+?examineExample（i）;
??????????end
?????else
??????????for?i?=?1:N
???????????????if?（alpha（i）?~=?0?&&?alpha（i）~=?boxconstraint（i））
????????????????????num_changed?=?num_changed+examineExample（i）;
???????????????end?
??????????end
?????end
?????
?????if?exam_all?==?1
??????????exam_all?=0;
?????elseif?num_changed?==?0?%當乘子全部檢查完后，標志位置1
??????????exam_all?=?1;
?????end??
?????
end
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%?檢查樣本。
????%?首先判斷i2是否滿足KKT條件，如果不滿足，則根據啟發式規則再選擇i1樣本，
????%?然后更新i1和i2的拉格朗日乘子。
????function?[flag]?=?examineExample（i2）
????????y2?=??Labels（i2）;
????????alpha2?=?alpha（i2）;
????????E2?=?wxpb（i2）?-?y2;
????????r2?=?E2?*?y2;
????????if?（r2??smo_opts.TolKKT?&&?alpha2?>?0）
????????????i1?=?selectSecondChoice（i2?E2）;
????????????if?i1?==?-1
????????????????i1?=?floor（1?+?rand（）?*?N）;?%?隨機選一個i1
????????????????while?i1?==?i2
????????????????????i1?=?floor（1?+?rand（）?*?N）;
????????????????end
????????????????flag?=?takeStep（i1i2）;
????????????else
????????????????flag?=?takeStep（i1i2）;
????????????end
????????else
????????????flag?=?0;
????????end
????end


?%?計算當前參數下第idx個樣本的函數輸出
????function?[svm_o]?=?wxpb（idx）
????????svm_o?=?0.0;
????????for?j?=?1?:?N
????????????svm_o?=?svm_o?+?alpha（j）?*?Labels（j）?*?kernel_func（training_data（j:）training_data（idx:））;
????????end
????????svm_o?=?svm_o?+?bias;
????end

????%?選定第二個變量i2后，根據max|E1-E2|的啟發式規則，選擇i1；
????%?如果沒有滿足條件的i1，返回-1.
????function?[i1]?=?selectSecondChoice（i2?E2）
????????i1?=?-1;
????????maxDelta?=?-1;
????????for?j?=?1?:?N
????????????if?j?~=?i2?&&?alpha（j）?~=?0?&&?alpha（j）?~=?boxconstraint（j）
????????????????Ej?=?wxpb（j）?-?Labels（j）;
????????????????if?abs（E2?-?Ej）?>?maxDelta
????????????????????i1?=?j;
????????????????????maxDelta?=?abs（E2?-?Ej）;
????????????????end
????????????end
????????end
????end

????%?根據選定的兩個變量i1i2，代入更新公式計算；
????%?最后還更新了偏移量offset，也就是y=wx+b中的b。
????function?[flag]?=?takeStep（i1?i2）
????????alpha1?=?alpha（i1）;
????????y1?=??Labels（i1）;
????????E1?=?wxpb（i1）?-?y1;
????????alpha2?=?alpha（i2）;
????????y2?=??Labels（i2）;
????????E2?=?wxpb（i2）?-?y2;
????????s?=?y1?*?y2;
????????if?y1?~=?y2
????????????L?=?max（0?alpha2?-?alpha1）;
????????

?屬性????????????大小?????日期????時間???名稱
-----------?---------??----------?-----??----
?????文件????????4654??2018-05-07?09:19??my_seqminopt.m
?????文件?????????825??2018-05-01?10:39??my_svm.m
?????文件???????10425??2018-05-01?16:29??my_svmtrain.m