addpath('./helper/');
workdir = './matfiles/';
heat_savedir = [workdir 'heat_kernels_q/'];
exp_savedir = [workdir 'experiments/svm/'];

experiments  = {{'C11'} {'E11'} {'G154'} {'M11'}};
%experiments = {{'C152'} {'E212'} {'GPOL'} {'GDIP'} {'M141'} {'M131'}};

% --- Prep For CV -------------------------------------------------------------
%load([workdir 'topics.mat']);
%load([workdir 'member.7k.mat']);
%maxNumCachedPerClass = 2500;
%numFolds             = 40;
%numTrainPerClass     = 100;
%numTestPerClass      = 200;
%
%for ii=1:length(experiments)
%	c1labels = experiments{ii};
%	% class1 will have label '1' while class2 will have label '-1'
%	[ids trn_samps tst_samps trn_zs tst_zs] = prepOneVsRest( c1labels, topics, member, ...
%		   maxNumCachedPerClass, numFolds, numTrainPerClass, numTestPerClass );
%
%	% shape c1labels into single string
%	labels = [c1labels;repmat({'_'},1,length(c1labels)-1) {''}];
%	labels = cell2mat(reshape(labels,1,2*numel(c1labels)));
%
%	save([exp_savedir labels 'vsAll.7k.mat'], ...
%		'maxNumCachedPerClass','numFolds','numTrainPerClass','numTestPerClass', ...
%		'c1labels','ids', 'trn_samps', 'tst_samps', 'trn_zs', 'tst_zs');
%end
%clear_except workdir heat_savedir exp_savedir experiments;




% --- Perform experiments -----------------------------------------------------
cc=[1.00 0.50 0.25];
tt=[0.96 1.12 1.28 1.44 1.60 1.76];
cut = 2000;

for ii=1:length(experiments)
	c1labels = experiments{ii};
	labels = [c1labels;repmat({'_'},1,length(c1labels)-1) {''}];
	labels = cell2mat(reshape(labels,1,2*numel(c1labels)));
	
	% retrieve folds and data
	load([exp_savedir labels 'vsAll.7k.mat']);
	load([workdir 'tf.7k.mat']);
	tf = tf(:,ids); % prune unused documents (cols)

	% prepare for feature selection
	[df I] = sort(sum(tf,2));
	Jsub = I((end-cut+1):end); % jvd: i.e., the words with highest doc freq get removed
	Jind = sub2ind([size(tf,1) size(tf,1)],Jsub,Jsub);

	fprintf('%s vs. All\n\tCV translated linear kernel (with feature selection):\n', labels);
	tacc = zeros(length(cc),length(tt),numFolds);
	for jj=1:length(cc)
		c = cc(jj);
		for kk=1:length(tt)
			t = tt(kk);
			
			H_file = [heat_savedir 'Hc' num2str(100*c) 't' num2str(100*t) '.7k.mat'];
			fprintf('\t\t%s\n',H_file);
			load(H_file);
			
			% --- Optional Feature selection (on T) ---------------------------
			H(Jsub,:)=0;H(:,Jsub)=0;H(Jind)=1;
			T = normalizeRows(H);	% jvd: make sure normalize last!

			% jvd: the linear kernel is simply:
			B   = tf' * T*T' * tf;
			TAcc = svmCVOneVsRest( trn_samps, tst_samps, trn_zs, tst_zs, tf, ...
						numFolds, '', 'tranlin', B);
			tacc(jj,kk,:) = TAcc;
		end
	end


	% --- Standard Linear Kernel ----------------------------------------------
	fprintf('CV Linear Kernel\n');
	acc = svmCVOneVsRest( trn_samps, tst_samps, trn_zs, tst_zs, tf, ...
	            numFolds, '', 'stndlin');
	save([exp_savedir 'acc_' labels '_lin_cut' num2str(cut/1000) 'k.7k.mat'],'cc','tt','cut','acc','tacc');
end