function [f,thr] = simpleroc(netout,truelab)%SIMPLEROC Basic receiver-operating characteristic curve%%      F = SIMPLEROC(NETOUT,TRUELAB)%% Compute the ROC curve for the network output NETOUT, given the true% labels TRUELAB. TRUELAB should contain 0-1 values, where 0 is the% correct output if (netout < some_threshold) and 1 when the% (netout > some_threshold). Output F contains [FNr FPr].%% This version returns a vector of the same length as NETOUT and% TRUELAB. Maybe this will be shortened/subsampled in the future.%% See also: dd_roc, plotroc% Copyright: D.M.J. Tax, D.M.J.Tax@prtools.org% Faculty EWI, Delft University of Technology% P.O. Box 5031, 2600 GA Delft, The Netherlands% Check the size of the netout vector:if size(netout,2)~=1	netout = netout';	if size(netout,1)~=1		error('Please make netout a column vector');	endend% So all sizes become:n = size(netout,1);n_t = sum(truelab);n_o = n - n_t;% Sort the network output:[sout,I] = sort(netout);% and reorder the true-labels accordingly:slab = truelab(I);% Make the index arrays for the target and outlier objects:slabt = slab;slabo = slab-1;% Check if there are identical outputs, and ...[uout,dummy,J] = unique(sout);% Change the slab such that the identical values are on top of each otherif size(uout,1)<n	%warning('dd_tools:NoUniqueROC',...	%	'There are identical values in NETOUT, the ROC is not uniquely defined.');	n = max(J);	slabt2 = zeros(n,1);	slabo2 = zeros(n,1);	for j=1:n  % count how many of each of the values occur, and store		% the approprate number in slabt2 and slabo2:		Jj = find(J==j);		slabt2(j) = sum(slabt(Jj));		slabo2(j) = sum(slabo(Jj));	end	slabt = slabt2;	slabo = slabo2;end% Faster version:	% number of 1's left from thresh.	% number of 0's right from thresh.f = [cumsum(slabt) (n_o+cumsum(slabo))];% fix the beginning of the curve to (0,n_o)f = [0 n_o; f];% ... and normalize:f = f./repmat([n_t n_o],n+1,1);% On request, also the thresholds are returned:if nargout>1	thr = uout;endreturn