Daniel@0: function [C,rate]=confmat(Y,T)
Daniel@0: %CONFMAT Compute a confusion matrix.
Daniel@0: %
Daniel@0: %	Description
Daniel@0: %	[C, RATE] = CONFMAT(Y, T) computes the confusion matrix C and
Daniel@0: %	classification performance RATE for the predictions mat{y} compared
Daniel@0: %	with the targets T.  The data is assumed to be in a 1-of-N encoding,
Daniel@0: %	unless there is just one column, when it is assumed to be a 2 class
Daniel@0: %	problem with a 0-1 encoding.  Each row of Y and T corresponds to a
Daniel@0: %	single example.
Daniel@0: %
Daniel@0: %	In the confusion matrix, the rows represent the true classes and the
Daniel@0: %	columns the predicted classes.  The vector RATE has two entries: the
Daniel@0: %	percentage of correct classifications and the total number of correct
Daniel@0: %	classifications.
Daniel@0: %
Daniel@0: %	See also
Daniel@0: %	CONFFIG, DEMTRAIN
Daniel@0: %
Daniel@0: 
Daniel@0: %	Copyright (c) Ian T Nabney (1996-2001)
Daniel@0: 
Daniel@0: [n c]=size(Y);
Daniel@0: [n2 c2]=size(T);
Daniel@0: 
Daniel@0: if n~=n2 | c~=c2
Daniel@0:   error('Outputs and targets are different sizes')
Daniel@0: end
Daniel@0: 
Daniel@0: if c > 1
Daniel@0:   % Find the winning class assuming 1-of-N encoding
Daniel@0:   [maximum Yclass] = max(Y', [], 1);
Daniel@0: 
Daniel@0:   TL=[1:c]*T';
Daniel@0: else
Daniel@0:   % Assume two classes with 0-1 encoding
Daniel@0:   c = 2;
Daniel@0:   class2 = find(T > 0.5);
Daniel@0:   TL = ones(n, 1);
Daniel@0:   TL(class2) = 2;
Daniel@0:   class2 = find(Y > 0.5);
Daniel@0:   Yclass = ones(n, 1);
Daniel@0:   Yclass(class2) = 2;
Daniel@0: end
Daniel@0: 
Daniel@0: % Compute 
Daniel@0: correct = (Yclass==TL);
Daniel@0: total=sum(sum(correct));
Daniel@0: rate=[total*100/n total];
Daniel@0: 
Daniel@0: C=zeros(c,c);
Daniel@0: for i=1:c
Daniel@0:   for j=1:c
Daniel@0:     C(i,j) = sum((Yclass==j).*(TL==i));
Daniel@0:   end
Daniel@0: end