class_decision.m

matlab工具箱的应用演示程序,

%	Program for decision.
%	Input is matrix n_class*n_patterns which is generated by NN.
%
%	The decision: for each column find values of maximal and next maximal element.
%	If their difference -'margin' is greater than treshold, decide index (row) of the element.
%	Otherwise, decide 0 (i.e., "I cannot decide").
%	Final result is COLUMN vector.
%
%	N.B When treshold is 0, we decide EVERYTHING (that's why we have <treshold...)
%
%	INPUTS:
%
%			results -  n_class*n_pattern matrix (output from NN)
%			treshold - if margin is smaller than treshold, there is no decision, i.e. decision is 0
%						  Treshold is either >0, or =0 (this is "normal" decision);
%	OUTPUTS:
%
%			decisions    column-vector of decided classes
%
% [decisions]=class_decision(results,treshold)

function  [decisions]=class_decision(results,treshold)

 [maxX,indeces,second_maxX,second_indeces]=second_max(results);

 decisions(find(maxX-second_maxX<treshold))=0;
 decided_positions=find(maxX-second_maxX>=treshold);
 decisions(decided_positions)=indeces(decided_positions);
 decisions=decisions';  %returns column vector

return;