sfs.m

次优搜索算法中常用的FSS算法

function [output_array, Subset, Cost_subsets, Cost,Same_subset] = SFS(X, Y,  new_size,data)
selectioncriteria='NN';

[n m] = size(X); % n-rows m-features

   j=1;
% Find the cost criterion for each feature

for(loop=1:new_size) 

    loop
    if(loop == 1)
        
        cost(1:m,1) = 0; % initialise
        cost(1:m,2) = [1:m]';
   
    else
        
        % Form new feat vect combos of winner and remaining features
    
        cost = zeros(K-1,L+1); %reinitialise
        lp=1;
    
        for(i=1:m)
        
            if(isempty(find(winner == i)) == 1) % i is not equal to any of the numbers made up by winner ie [2 3 4]
        
                cost(lp,1)=0; % Reinitialise cost value
            
                cost(lp,2:L+1) = [winner i];
                lp=lp+1;
            end
        end
    
    end
    
    
    %cost = zeros(m+(loop-1),2+(loop-1));
    
    [K L] = size(cost); % find length of feat vector

    for(i=1:K) % m-feature subsets 9,8,7,6-> (x1), (x1 x2), (x1 x2 x3), etc...

        % cost array = cost value, feature sequence x No feat combo's
        
        % Form feature test vector, form different combinations (x1), (x1 x2), (x1 x2 x3), etc...
       
        subset=cost(i,2:L);

 %       cost(i,1) = rand();
    cost(i,1) = fs_knn(subset,X,Y,data);
    end

 
    % perform quicksort of measure column

    B = sortrows(cost,1);

    winner =  B(K,2:L); % feat combo of winner
    
    %--------my code for same performance for different subset
   nSame=0;
    for(i=1:K)   %count the number of subset
       if(B(i,1)==B(K,1))
          nSame=nSame+1;
         Same_subset(j,1:L-1)=B(i,2:L); %store  same performance for feature subset 
        
          subpop_size=1;
       T=1;
       pc1=1;                     
      
        j=j+1;
       
       end      
      
    end
    
     if(nSame==1)
           j=j-1;%only one subset,no storage
           
     end
       
     if(nSame>1)
        % [p q]=size(Same_subset)
          for(i=1:nSame)
           k=write_file(Same_subset(j-nSame+i-1,:),subpop_size,T,pc1);%only  subset
           [winner2, Cost2, retfeat] = addfeat(X, Y, selectioncriteria, Same_subset(j-nSame+i-1,:));
           one_feature(i,1)=Cost2;
           one_feature(i,2)=retfeat;
          end
          one_feature;
         [max_subset  index] = sortrows(one_feature,1);
          one_feature(nSame,2)
          
          mySubset =  B(K-3+index(nSame),2:L); % max J(X+) 
            winner =  B(K-3+index(nSame),2:L); % modify the winner ,choose the search direction 
     end
             
   
     
 %    [max_subset, Subset1, Cost_subsets1, Cost1,Same_subset1] = search_max(X, Y, Same_subset);
     
  %  j=loop+j+1;
    %-----------------------end-----------
   
    
    
    % Ammend Subset & Cost to Stored subsets

    Subset(L-1,1:L-1)=B(K,2:L);
    
    Cost_subsets(L-1)=B(K,1); 
    
    end


    
%Same_subset=Same_subset;    
    
Cost = B(K,1); % Lst! winners Cost

% return optimal feature vector combo

output_array = winner;

return; % End function