👨🎓个人主页：研学社的博客
💥💥💞💞欢迎来到本博客❤️❤️💥💥
🏆博主优势：🌞🌞🌞博客内容尽量做到思维缜密，逻辑清晰，为了方便读者。
⛳️座右铭：行百里者，半于九十。
📋📋📋本文目录如下：🎁🎁🎁
💥1 概述
特征选择（特征子集选择）问题是各个领域重要的预处理阶段之一。在真实的数据集中，存在许多不相关、误导性和冗余的特征，这些特征是无用的。主要特征可以通过特征选择技术提取。特征选择属于 NP 难题类;因此，元启发式算法对于解决问题很有用。引入一种新的二元ABC称为二元多邻域人工蜂群（BMNABC），以增强ABC阶段的勘探和开发能力。BMNABC 在第一阶段和第二阶段使用新的概率函数应用近邻域和远邻域信息。在第三阶段，对那些在前一阶段没有改进的解决方案进行了比标准ABC更有针对性的搜索。该算法可以与包装器方法结合使用，以获得最佳结果。
📚2 运行结果
部分代码：
function [BestChart,Error,BestFitness,BestPosition]= BMNABC(FoodNumber,maxCycle,D,fobj,Limit)
global SzW
rand('state',sum(100*clock)) %#ok
%reset trial counters
trial=zeros(1,FoodNumber);
rmin=0;rmax=1;
Vmax=6;
BestChart=[];
%%
%/FoodNumber:The number of food sources equals the half of the colony size*/
Foods=initialization(FoodNumber,D,1,0)>0.5;
for i=1:FoodNumber
ObjVal(i) =feval(fobj,Foods(i,:));
end
BestInd=find(ObjVal==min(ObjVal));
BestInd=BestInd(end);
BestFitness=ObjVal(BestInd);
BestPosition=Foods(BestInd,:);
x=BestPosition;
Acc=(SzW*sum(x)/length(x)-BestFitness)/(1-SzW)+1; % This relation is computed based on FitnessValue=(1-SzW)*(1-cp.CorrectRate)+SzW*sum(x)/(length(x));
% The feature subset is based on two criteria: the maximum classi?cation accuracy (minimum error rate) and the minimal number of selected features.
%Fitness = a* ClassificationError+ (1-a) NumberOfSelectedFeatures/TotalFeatures
Error=1-Acc;
BestChart=[BestChart,Error];
fprintf('Iteration:%3d\tAccuracy:%7.4f\tFitness:%7.4f\t#Selected Features:%d\n',1,((1-Error)*100),BestFitness,(sum(BestPosition,2)));
gindex=BestInd;
pbest=Foods;
pbestval=ObjVal;
iter=2;
while ((iter <=maxCycle)),
r= rmax-iter*(rmax-rmin)/maxCycle;
%%
%%%%%%%%% EMPLOYED BEE PHASE %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
for i=1:(FoodNumber)
neigh=[]; d=[]; dd=[];
w=0;
for k=1:FoodNumber
if k~=i
w=w+1;
d(w)=sum(Foods(i,:)~=Foods(k,:)); %Hamming distance.
dd(w)=k;
end
end
Meand=mean(d);
Maxd=max(d);
u=1;
for z=1:w
if (d(z)<=Maxd && d(z)>=r*Meand)
neigh(u)=dd(z);
u=u+1;
end
end
fi=rand(length(neigh),D);
FIP=sum(fi.*pbest(neigh,:))./sum(length(neigh));
sol=Foods(i,:);
V(i,:)=rand(1,D).*(Foods(i,:)-FIP);
V(i,:)=max(min(V(i,:),Vmax),-Vmax);
aa=abs(tanh(V(i,:)));
AA=1-exp(-trial(i)/(iter*maxCycle));
S=AA+aa;
temp=rand(1,D)<S;
moving=find(temp==1);
sol(moving)=~sol(moving);
ObjValSol=feval(fobj,sol);
if (ObjValSol<pbestval(i))
pbest(i,:)=sol;
pbestval(i)=ObjValSol;
end
if (ObjValSol<ObjVal(i))
Foods(i,:)=sol;
ObjVal(i)=ObjValSol;
trial(i)=0;
else
trial(i)=trial(i)+1; %/*if the solution i can not be improved, increase its trial counter*/
end;
end;
%%
%%%%%%%%%%%%%%%%%%%%%%%% ONLOOKER BEE PHASE %%%%%%%%%%%%%%%%%%%%%%%%%%%
i=0;
while(i<FoodNumber)
i=i+1;
neigh=[]; d=[];dd=[];
w=0;
for k=1:FoodNumber
if k~=i
w=w+1;
d(w)=sum(Foods(i,:)~=Foods(k,:));
dd(w)=k;
end
end
Meand=mean(d);
u=1;
for z=1:w
if (d(z)<=Meand)
neigh(u)=dd(z);
u=u+1;
end
end
[tmp,tmpid]=min(ObjVal(neigh));
nn=neigh(tmpid);
sol=Foods(i,:);
V(i,:)=rand(1,D).*(Foods(i,:)-Foods(nn,:));
V(i,:)=max(min(V(i,:),Vmax),-Vmax);
aa=abs(tanh(V(i,:)));
AA=1-exp(-trial(i)/(iter*maxCycle));
S=AA+aa;
temp=rand(1,D)<S;
moving=find(temp==1);
sol(moving)=~sol(moving);
ObjValSol=feval(fobj,sol);
if (ObjValSol<pbestval(i))
pbest(i,:)=sol;
pbestval(i)=ObjValSol;
end
% /*a greedy selection is applied between the current solution i and its mutant*/
if (ObjValSol<ObjVal(i)) %/*If the mutant solution is better than the current solution i, replace the solution with the mutant and reset the trial counter of solution i*/
Foods(i,:)=sol;
ObjVal(i)=ObjValSol;
trial(i)=0;
else
trial(i)=trial(i)+1; %/*if the solution i can not be improved, increase its trial counter*/
end;
end
%%
%/*The best food source is memorized*/
ind=find(ObjVal==min(ObjVal));
ind=ind(end);
if (ObjVal(ind)<BestFitness)
BestFitness=ObjVal(ind);
BestPosition=Foods(ind,:);
x=BestPosition;
Acc=(SzW*sum(x)/length(x)-BestFitness)/(1-SzW)+1;
Error=1-Acc;
gindex=ind;
success=1;
else
success=0;
end;
%%
%%%%%%%%%%%% SCOUT BEE PHASE %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%/*determine the food sources whose trial counter exceeds the "Limit" value.
%In Basic ABC, only one scout is allowed to occur in each cycle*/
ind=find(trial==max(trial));
ind=ind(end);
if (trial(ind)>Limit)
sol=Foods(ind,:);
b=randi([1 FoodNumber]);
while ((b==ind)|| (b==gindex))
b=randi([1 FoodNumber]);
end
if (success==1)
sol=or(Foods(ind,:),BestPosition) ;
else
sol= or (xor(Foods(ind,:),BestPosition) ,or(Foods(ind,:),Foods(b,:)));
end
ObjValSol=feval(fobj,sol);
if (ObjValSol<pbestval(ind))
pbest(ind,:)=sol;
pbestval(ind)=ObjValSol;
end
if (ObjValSol<ObjVal(ind)) %/*If the mutant solution is better than the current solution i, replace the solution with the mutant and reset the trial counter of solution i*/
Foods(ind,:)=sol;
ObjVal(ind)=ObjValSol;
trial(ind)=0;
else
trial(ind)=trial(ind)+1; %/*if the solution i can not be improved, increase its trial counter*/
end;
end
BestChart=[BestChart,Error];
fprintf('Iteration:%3d\tAccuracy:%7.4f\tFitness:%7.4f\t#Selected Features:%d\n',iter,((1-Error)*100),BestFitness,(sum(BestPosition,2)));
iter=iter+1;
end
🎉3 参考文献
部分理论来源于网络，如有侵权请联系删除。
Zahra Beheshti(2018), BMNABC: Binary Multi-Neighborhood Artificial Bee Colony for High-Dimensional Discrete Optimization Problems, Cybernetics and Systems 49 (7-8), 452-474.
