calculateDecodePerformance.m (90f4202) - SVMCrossVal.git

calculateDecodePerformance.m

% function [decodePerformance rawTimecourse ] = calculateDecodePerformance(des,timeLineStart, timeLineEnd, decodeDuration, svmargs, conditionList, sessionList, voxelList, classList, labelMap,normalize)
function outputStruct = calculateDecodePerformance(inputStruct,SubjectID)

addpath 'libsvm-mat-2.88-1';

SINGLE = 1;

outputStruct = struct;

namehelper      = strcat('s',SubjectID);
des             = inputStruct.(namehelper).des;
timeLineStart   = inputStruct.frameShiftStart;
timeLineEnd     = inputStruct.frameShiftEnd;
decodeDuration  = inputStruct.decodeDuration;
svmargs         = inputStruct.svmargs;
sessionList     = inputStruct.sessionList;
voxelList       = inputStruct.(namehelper).voxelList;
% classList       = inputStruct.classList;
% labelMap        = inputStruct.labelMap;
smoothed       = inputStruct.smoothed;
globalStart     = inputStruct.psthStart;
globalEnd       = inputStruct.psthEnd;
baselineStart   = inputStruct.baselineStart;
baselineEnd     = inputStruct.baselineEnd;
eventList       = inputStruct.eventList;
labelMap        = inputStruct.labelMap;

minPerformance = inf;
maxPerformance = -inf;

%Pro Voxel PSTH TIMELINE berechnen.
        %   timeshift mit pst-timeline durchf�hren.
        % psth-timeline -25 bis +15 zu RES Onset.
        
%         eventList       = [9,11,13;10,12,14];
%         globalStart     = -25;
%         globalEnd       = 15;
%         baselineStart   = -22;
%         baselineEnd     = -20;
        
        
        for voxel = 1:size(voxelList,1)  % [[x;x],[y;y],[z;z]]
                extr  = calculateImageData(voxelList(voxel,:),des,smoothed);
                rawdata=cell2mat({extr.mean}); % Raw Data
                pst{voxel}  = calculatePST(des,globalStart,baselineStart,baselineEnd,globalEnd,eventList,rawdata,sessionList);
        end
        
        decodePerformance = [];

for timeShift   = timeLineStart:1:timeLineEnd
            frameStart  = floor(-globalStart+1+timeShift - 0.5*decodeDuration);
            frameEnd    = min(ceil(frameStart+decodeDuration + 0.5*decodeDuration),-globalStart+globalEnd);
            
            tmp =[];
            anyvoxel = 1;
            for pstConditionGroup = 1:size(pst{1,anyvoxel},2) 
                for dp = 1:size(pst{1,anyvoxel}{1,pstConditionGroup},1) % data point
                  row = getSVMLabel(labelMap,eventList(pstConditionGroup,1));
                    for voxel = 1:size(pst,2)
                        row = [row, pst{1,voxel}{1,pstConditionGroup}(dp,frameStart:frameEnd)]; % label,value,value
                    end
                tmp  = [tmp; row];
                end
            end 
        
            svmdata      = tmp(:,2:size(tmp,2));
            svmlabel     = tmp(:,1);
            
%             rndindex  = randperm(length(svmlabel));
%             svmdata   = svmdata(rndindex,:);
%             svmlabel  = svmlabel(rndindex);
            
            if SINGLE
                performance  = svmtrain(svmlabel, svmdata, svmargs);

minPerformance = min(minPerformance,performance);
                maxPerformance = max(maxPerformance,performance);

decodePerformance = [decodePerformance; performance];
            end
            
        end
        
        outputStruct.decodePerformance  = decodePerformance;
        outputStruct.svmdata            = svmdata;
        outputStruct.svmlabel           = svmlabel;
        outputStruct.rawTimeCourse      = pst;
        outputStruct.minPerformance     = minPerformance;
        outputStruct.maxPerformance     = maxPerformance;
end

SVMCrossVal.git

plot optimiert. normalisierung pst, zwischenversion UI Defaults verbessert