SVMCrossVal.git

@@ -19,4 +19,7 @@ SVMCROSSVAL_VOXEL_SELECTION_MODE_DEF.roiImage     = 'use ROI image by pop-up ima

 global SVMCROSSVAL_SUBJECT_PREFIX;

 % internally used to prefix subject-ids starting with numbers.

 SVMCROSSVAL_SUBJECT_PREFIX                        = 'subject';

+

+global SVMCROSSVAL_SUBJECTSTRUCT_NAME;

+SVMCROSSVAL_SUBJECTSTRUCT_NAME = 'subjectStruct';

 end

\ No newline at end of file

@@ -1,74 +1,23 @@

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

-function outputStruct = calculateDecodePerformance(timeline,inputStruct,subjectParams)

-

-global CROSSVAL_METHOD_DEF;

-

-

-addpath 'libsvm-mat-2.88-1';

-

-% CROSSVAL_METHOD_DEF = inputStruct.CROSSVAL_METHOD_DEF;

-METHOD              = inputStruct.CROSSVAL_METHOD;

-

-RANDOMIZE_DATAPOINTS = inputStruct.RANDOMIZE;

-

-% SubjectID       = subjectParams.SubjectID;

-% namehelper      = subjectParams.namehelper;

-voxelList       = subjectParams.voxelList;

-des             = subjectParams.des;

-

+function outputStruct = calculateDecodePerformance(timeline,subjectStruct,model)

 outputStruct = struct;

+RANDOMIZE_DATAPOINTS = 0;

-svmargs         = inputStruct.svmargs;

-sessionList     = inputStruct.sessionList;

-% classList       = inputStruct.classList;

-% labelMap        = inputStruct.labelMap;

 eventList       = inputStruct.eventList;

 timeLineStart   = timeline.frameShiftStart;

 timeLineEnd     = timeline.frameShiftEnd;

-% decodeDuration  = timeline.decodeDuration;

-% globalStart     = timeline.psthStart;

-% globalEnd       = timeline.psthEnd;

-% baselineStart   = timeline.baselineStart;

-% baselineEnd     = timeline.baselineEnd;

-

-

-minPerformance = inf;

-maxPerformance = -inf;

-subjectDir = '';

-sessionDirList = sessionList2DirList(sessionList) ;

-mask = '^fandersen.*img$';

-imageFiles = getImageFileList(subjectDir,sessionDirList,mask);

-

-

-disp('press key');

-pause

-

-extr = calculateImageData(imageFiles,voxelList);

-

-nVoxel = size(voxelList,1);

-

-calculatePstOpts = struct;

-calculatePstOpts.des = des;

-calculatePstOpts.eventList = eventList;

-calculatePstOpts.sessionList = sessionList;

+% for iVoxel = 1:nVoxel

+%     rawdata = [];

+%     for iImage = 1:length(extr);

+%         tmp = extr(iImage);

+%         rawdata = [rawdata tmp.dat(iVoxel)];

+%     end

+%     pst{iVoxel} = calculatePST(timeline,calculatePstOpts,rawdata);

+% end

-for iVoxel = 1:nVoxel

-    rawdata = [];

-    for iImage = 1:length(extr);

-        tmp = extr(iImage);

-        rawdata = [rawdata tmp.dat(iVoxel)];

-    end

-    pst{iVoxel} = calculatePST(timeline,calculatePstOpts,rawdata);

-end

-%         for voxel = 1:size(voxelList,1)  % [[x;x],[y;y],[z;z]]

-%             extr        = calculateImageData(imageFiles,voxelList(voxel,:));

-%             rawdata     = cell2mat({extr.mean}); % Raw Data

-%             pst{voxel}  = calculatePST(des,globalStart,baselineStart,baselineEnd,globalEnd,eventList,rawdata,sessionList);

-%         end

 timePointArgs.pst = pst;

@@ -89,45 +38,8 @@ for index = 1:timeLineEnd-timeLineStart+1

         svmlabel  = svmlabel(rndindex);

     end

-    switch METHOD;

-        case CROSSVAL_METHOD_DEF.svmcrossval

-

-            performance  = svmtrain(svmlabel, svmdata, svmargs);

-

-            minPerformance = min(minPerformance,performance);

-            maxPerformance = max(maxPerformance,performance);

-

-            decodePerformance = [decodePerformance; performance];

-

-        case CROSSVAL_METHOD_DEF.classPerformance

-

-            newsvmopt = killCrossvalOpt(svmargs);

-

-            model = svmtrain(svmlabel,svmdata,newsvmopt);

-            classperformance = [];

-            for class = unique(svmlabel)';

-

-                filterindex = find(class == svmlabel);

-                testing_label = svmlabel(filterindex);

-                testing_data  = svmdata(filterindex);

-                [plabel accuracy dvalue] = svmpredict(testing_label,testing_data,model,'');

-

-                classperformance = [classperformance accuracy(1)];

-            end

-            decodePerformance = [decodePerformance; classperformance];

-

-        case CROSSVAL_METHOD_DEF.somTraining

+    decodePerformance = [decodePerformance; svm_single_crossval(svmlabel,svmdata,svmopts)];

-            display('SOM TRAINING');

-            addpath 'somtoolbox2';

-            sD = som_data_struct(svmdata,'label',num2str(svmlabel));

-            assignin('base','sD',sD);

-            sM = som_make(sD,'msize', [3 3],'lattice', 'hexa');

-

-            assignin('base','sD',sD);

-            assignin('base','sM',sM);

-            display('type ''figure'' before visualisation');

-    end

 end

@@ -139,19 +51,5 @@ outputStruct.minPerformance     = minPerformance;

 outputStruct.maxPerformance     = maxPerformance;

 end

-function opts = killCrossvalOpt(svmopt)

-opts = '';

-idx1 = 1;

-for idx2=strfind(svmopt,' -')

-    if idx1 ~= strfind(svmopt,' -v')

-        opts = strcat(opts,svmopt(idx1:idx2));

-    end

-    idx1=idx2;

-    if idx2==max(strfind(svmopt,' -'))

-        opts = strcat(opts,svmopt(idx2:end));

-    end

-end

-end

-

@@ -0,0 +1,10 @@

+function svmargs = getSvmArgs(model,single_run)

+

+    svmargs = get(model.txtSVMopts,'String');

+

+if single_run

+    svmargs = [svmargs ' -v ' num2str(getSvmNFold(model))];

+end

+

+

+end

\ No newline at end of file

@@ -0,0 +1,3 @@

+function nfold = getSvmNFold(model)

+nfold = str2double(get(model.txtSVMnfold,'String'));

+end

\ No newline at end of file

@@ -38,13 +38,20 @@ switch task

         roiargs.sessionList = 1:3;

         roiargs.eventList   = classDef.eventMatrix;

-        

-        assignin('base','roiargs',roiargs);

-        

         runROIImageMaskMode(roiargs);

     case 'FBS'

         disp('FBS')

+        

+    case 'SVM'

+        disp('classify with svm');

+        

+    case 'X-SVM'

+        

+    case 'SOM'

+

+    case 'X-SOM'

+        

 end

 % disp('warings restored');

@@ -1,4 +1,7 @@

 function runCoordTable(args)

+

+global SVMCROSSVAL_SUBJECTSTRUCT_NAME;

+

     disp('run coord table')

     subjects = args.subjects;

@@ -37,33 +40,5 @@ function runCoordTable(args)

        disp(sprintf('done %g // %g',s,nSubjects));

     end

-    assignin('base','subjectStruct',subjectStruct);

+    assignin('base',SVMCROSSVAL_SUBJECTSTRUCT_NAME,subjectStruct);

 end

-

-

-

-

-% 

-% %         decode = claculateMultiSubjectDecodePerformance(timelineParams,calculateParams,paramModel);

-% 

-%         display('Finished calculations.');

-%         display('Plotting...');

-% 

-%         plotParams                   = struct;

-%         

-% %         plotParams.SVMCROSSVAL_CROSSVAL_METHOD_DEF = SVMCROSSVAL_CROSSVAL_METHOD_DEF;

-%         plotParams.CROSSVAL_METHOD     = calculateParams.CROSSVAL_METHOD;

-%         

-%         plotParams.nClasses          = length(calculateParams.classList);

-% 

-%         plotParams.decodePerformance = decode.decodePerformance;

-%         plotParams.rawTimeCourse     = decode.rawTimeCourse;

-%         plotParams.SubjectID         = subjectSelection;

-%         plotParams.smoothed          = boolToYesNoString(calculateParams.smoothed);

-% 

-%         assignin('base','plotParams',plotParams);

-% %         plotDecodePerformance(params.psthStart,params.psthEnd,params.nClasses,decode.decodeTable,params.frameShiftStart,params.frameShiftEnd,decode.rawTimeCourse);

-%         plotDecodePerformance(timelineParams,plotParams);

-%             

-%         display('all done.');

-% 

@@ -1,5 +1,7 @@

 function runROIImageMaskMode(args)

+global SVMCROSSVAL_SUBJECTSTRUCT_NAME;

+

 subjects = args.subjects;

 nSubjects = size(subjects);

@@ -47,7 +49,7 @@ for s = 1:nSubjects

     disp('done');

 end

-assignin('base','subjectStruct',subjectStruct);

+assignin('base',SVMCROSSVAL_SUBJECTSTRUCT_NAME,subjectStruct);

 end

@@ -0,0 +1,16 @@

+function decodePerformance = svm_crossval(svmlabel,svmdata,svmopts)

+addpath 'libsvm-mat-2.88-1';

+

+svmmodel = svmtrain(svmlabel,svmdata,svmopts);

+classperformance = [];

+for class = unique(svmlabel)';

+

+    filterindex = find(class == svmlabel);

+    testing_label = svmlabel(filterindex);

+    testing_data  = svmdata(filterindex);

+    [plabel accuracy dvalue] = svmpredict(testing_label,testing_data,svmmodel,'');

+

+    classperformance = [classperformance accuracy(1)];

+end

+decodePerformance = [decodePerformance; classperformance];

+end

\ No newline at end of file

@@ -0,0 +1,4 @@

+function performance = svm_single_crossval(svmlabel,svmdata,svmopts)

+addpath 'libsvm-mat-2.88-1';

+performance  = svmtrain(svmlabel, svmdata, svmopts);

+end

\ No newline at end of file

@@ -0,0 +1,12 @@

+function [sD sM] = train_som(svmlabel, svmdata, somOptions)

+

+display('SOM TRAINING');

+addpath 'somtoolbox2';

+sD = som_data_struct(svmdata,'label',num2str(svmlabel));

+assignin('base','sD',sD);

+sM = som_make(sD,'msize', [3 3],'lattice', 'hexa');

+

+assignin('base','sD',sD);

+assignin('base','sM',sM);

+display('type ''figure'' before visualisation');

+end

\ No newline at end of file

@@ -91,7 +91,7 @@ function model = mcb_load(src,evnt,model)

 disp('LOAD');

 [file path] = uigetfile('*.mat','Load Params ...',model.baseDir);

 l = load(fullfile(path,file));

-assignin('base','l',l);

+% assignin('base','l',l);

 model = setTimeLineParams(model,l.timeLine);

 model = setClassDefString(model,l.classDefString);

 model = setCoordDefString(model,l.coordDefString);

@@ -319,12 +319,10 @@ function model = createFirstStepPanel(model,parent,DEFAULT)

             'Units','normalized','Position',[0.66 0 0.33 1]);

         set(btnRunButton3,'Callback',{@cbRunPreprocessing,model,'ROI'}); % set here, because of model.    

         set(btnRunButton3,'Enable','on');

-        

-        assignin('base','model',model);

 end

 function cbRunPreprocessing(src,evnt,model,task)

-main(model,task)

+main(model,task);

 end

 function label = createLabel(parent,  pos, labelText)