roc.m

模式识别 MATLAB 的工具箱,比较实用,包括SVM,ICA,PCA,NN等等模式识别算法.

%ROC Receiver-operator curve
% 
% 	e = roc(D,k)
% 
% Computes k points of the receiver-operator curve of the classifier 
% W for the labeled data set D, which is typically the result of
% D = A*W*classc, The curve is computed for k thresholds of the 
% aposteriori probabilities stored in D. The resulting error 
% frequencies for the two classes are stored in the two columns of 
% e, which may conveniently be plotted by plot2. Default k = 100
% 
% See also datasets, mappings, reject, plot2

% Copyright: R.P.W. Duin, [e-mail]
% Faculty of Applied Physics, Delft University of Technology
% P.O. Box 5046, 2600 GA Delft, The Netherlands

function e = roc(A,k);
if nargin < 2, k = 100; end
if nargin == 0 | isempty(A)
	e = mapping('roc','fixed',k);
	return
end
[nlab,lablist,m,c,cc,p] = dataset(A);
e1 = []; e2 = [];
n = size(A,1);
step = ceil(n/k);
if step ~=1
	k = k-1;
end
for i=1:c
	d = sort(+A(:,i));
	n = length(d);
	step = ceil(n/k);
	nk = step*k-n;
	nk1 = floor(nk/2);
	nk2 = ceil(nk/2);
	d = [repmat(d(1),nk1,1);d;repmat(d(end),nk2,1)];
	d = reshape(d,step,k);
	if step ~= 1
		d = [d(1,:) d(end)];
	end
	n = length(d);
	labo = repmat(A(:,i),1,n) > repmat(d,m,1);
	q = repmat(nlab==i,1,n) ~= labo;
	e1 = [e1; mean(q(find(nlab==i),:),1)];
	e2 = [e2; mean(q(find(nlab~=i),:),1)];
end
e = [mean(e1,1); mean(e2,1)];