x_valid.m

The neuro-fuzzy software for identification and data analysis has been implemented in the MATLAB lan

function  [errors]=x_valid(data,comp_min,comp_max)


%X_VALID		Cross Validation
% 
%  [errors]=x_valid(data,comp_min,comp_max)
%
%	errors[no_sets,comp_max-comp_min+1]	
%			
%	data[no_data,no_var+1]		the whole data avaliable
%	comp_min[1,1]
%	comp_max[1,1]
%
%	x_valid executes, for each degree of complexity between 
%	comp_min and comp_max, the (no_sets)Cross_Validation 
%	for a model of the type specified in the global variable "model".
%	It returns a matrix whose k-th column is composed by 
%	the errors obtained on the different training sets, 
%	for the complexity (k-1+comp_min).
%
%	The following global variable are required:
%		model
%		no_sets
%		fid_h
%		file_e

%______________________________________________________10 April 1996___

global model
global no_sets

global fid_h
global file_e


no_data=size(data,1);

%sets_size[no_sets,1] is a vector containing the size of each set
sets_size=fix(no_data/no_sets)*ones(no_sets,1);
for i=1:rem(no_data,no_sets)
	sets_size(i,1)=sets_size(i,1)+1;
end

%Permutation of the lines of the matrix data
permutation=randperm(no_data);
data_perm=zeros(size(data));
for i=1:no_data
	data_perm(i,:)=data(permutation(i),:);
end
data=data_perm;
			
select_mat=zeros(no_data,no_sets);
line_counter=1;
for i=1:no_sets
	for j=1:sets_size(i)
		select_mat(line_counter,i)=1;
		line_counter=line_counter+1;
	end
end
%select_mat[tot_samples,no_sets]
%where no_sets is the number of 'folds' in a 'n-fold'-Cross_Validation:
%In the matrix select_mat, one and only one element per row
%equals one, all the others are zeros.
%If in the i-th row, the one is in the j-th column it means
%that the i-th sample belongs to the j-th test set
%and to all the training sets but the j-th.




%__Complexity cycle______________________________________

errors=zeros(no_sets,comp_max-comp_min+1);

for comp=comp_min:comp_max
		
	for s=1:no_sets
		fprintf(fid_h,'\nComplexity: %g on %g \t X-validation: %g on %g\n',comp,comp_max,s,no_sets);
		

		%creation of the current test and training sets
		temp=select_mat;
		temp(:,s)=zeros(size(temp,1),1);
		training_set=data(find(sum(temp')),:);
		test_set=data(find(select_mat(:,s)),:);
		[no_samples,c]=size(test_set);
		in_tst=test_set(:,1:c-1);
		out_tst=test_set(:,c);
	
		[out_hat]=eval([model,'(comp,training_set,in_tst)']); 
				
		RMSE=sqrt(sumsqr(out_tst-out_hat)/(no_samples-1));
		

		fprintf(fid_h,'TRUE RMSE: %g\n\n',RMSE);
		errors(s,comp-comp_min+1)=RMSE;
		eval(['save ', file_e ,' errors']);

	end
end