SVMCrossVal.git

@@ -1,6 +1,7 @@

 function outputStruct = calculateDecodePerformance(header,subjectStruct,svmopts)

 outputStruct = struct;

 RANDOMIZE_DATAPOINTS = header.svmrnd;

+NAN_AS_ZERO = header.nantozero;

 timeline = header.timeline;

 timeline.frameShiftStart = header.frameShift.frameShiftStart;

@@ -37,6 +38,11 @@ for index = 1:timeLineEnd-timeLineStart+1

         svmlabel  = svmlabel(rndindex);

     end

+    if NAN_AS_ZERO

+        svmdata(isnan(svmdata))=0;

+    end

+        

+    

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

@@ -122,17 +122,22 @@ switch task

         disp('SVM');

         svmopts    = getSvmArgs(model,1);

         header.svmrnd = getSvmRnd(model);

+        header.nantozero = 1;

         decode = calculateMultiSubjectDecodePerformance(header,data,svmopts);

         decode.header = header;

         assignin('base','decode',decode);

     case 'XSVM'

         disp('XSVM')

         svmopts  = getSvmArgs(model,0);

+        header.svmrnd = getSvmRnd(model);

+        header.nantozero = 1;

         decode = xsvm_subject_loop(header,data,svmopts);

         decode.header = header;

         assignin('base','decode',decode);

     case 'SOM'

         display('SOM');

+        somopts.rnd = 1;

+        somopts.nantozero = 1;

         somopts.size = [3 3];

         somopts.lattice = 'rect';

         somopts.nfold = 6;

@@ -141,6 +146,8 @@ switch task

         assignin('base','decode',decode);

     case 'XSOM'

         display('XSOM');

+        somopts.rnd = 1;

+        somopts.nantozero = 1;

         somopts.size = [3 3];

         somopts.lattice = 'rect';

         decode = som_xsubject_performance(header,data,somopts);

@@ -129,7 +129,7 @@ for s = 1:nSubjects

     for timeShiftIdx = 1:nSamplePoints

     % center timepoint && relative shift

     frameStartIdx  = floor(-globalStart+1+timeShiftIdx - 0.5*decodeDuration);

-    frameEndIdx    = min(ceil(frameStart+decodeDuration + 0.5*decodeDuration),-globalStart+globalEnd);

+    frameEndIdx    = min(ceil(frameStartIdx+decodeDuration + 0.5*decodeDuration),-globalStart+globalEnd);

         img3D = zeros(size(mask_image)); %output image prepare

@@ -3,7 +3,8 @@ function decode = som_subject_batch(header,subjectdata,somOpts)

 addpath 'somtoolbox2';

-RANDOMIZE_DATAPOINTS = 1;

+RANDOMIZE_DATAPOINTS = somOpts.rnd;

+NAN_AS_ZERO = somOpts.nantozero;

 decode = struct;

 decode.decodePerformance = [];

@@ -62,9 +63,18 @@ for subjectDataID = 1:nSubjects

             svm_train_label(chunkstart:chunkend) = []; %del test set

             svm_train_data(chunkstart:chunkend,:) = [];% del test set

-            [sD sM] = som_train(svm_train_label, svm_train_data, somOpts);

+            if NAN_AS_ZERO

+                svm_train_data(isnan(svm_train_data))=0;

+                svm_validation_data(isnan(svm_validation_data))=0;

+                display('NaN to 0');

+            end

+            if isempty(svm_train_data)

+                performance = 0;

+            else

+                [sD sM] = som_train(svm_train_label, svm_train_data, somOpts);

                 performance = som_decode(sM, svm_validation_data,svm_validation_label);

+            end

             cross_value = [cross_value, performance];

         end

@@ -6,7 +6,7 @@ som_lattice = somOptions.lattice;

 addpath 'somtoolbox2';

 sD = som_data_struct(svmdata,'labels',num2str(svmlabel));

-sM = som_make(sD,'msize', som_size,'lattice', som_lattice);

+sM = som_make(sD,'msize', som_size,'lattice', som_lattice, 'tracking', 1, 'init', 'lininit');

 sM = som_autolabel(sM,sD,'vote');

 end

\ No newline at end of file

@@ -8,7 +8,8 @@ if(nSubjects < 2)

     error('SVMCrossVal:somXSubjectPerformance:tooFewSubjects','You need at least 2 Subjects in this Across-Subject analysis!');

 end

-RANDOMIZE_DATAPOINTS = 1;

+RANDOMIZE_DATAPOINTS = somOpts.rnd;

+NAN_AS_ZERO = somOpts.nantozero;

 decode = struct;

 decode.decodePerformance = [];

@@ -65,9 +66,17 @@ for timeIndex = 1:timeLineEnd-timeLineStart+1

             svm_train_label  = svm_train_label(rndindex);

         end

+        if NAN_AS_ZERO

+            svm_train_data(isnan(svm_train_data))=0;

+        end

+        

+        if isempty(svm_train_data)

+            performance = 0;

+        else

             [sD sM] = som_train(svm_train_label, svm_train_data, somOpts);

             performance = som_decode(sM, svm_validation_data,svm_validation_label);

+        end

         cross_value = [cross_value performance];

     end

@@ -1,14 +1,14 @@

 function ui_main(varargin)

-DEFAULT.selectedSubject = 2;

+DEFAULT.selectedSubject = [2];

-DEFAULT.pststart        = -15;

-DEFAULT.pstend          = 40;

-DEFAULT.baselinestart   = -3;

-DEFAULT.baselineend     = -1;

+DEFAULT.pststart        = -5;

+DEFAULT.pstend          = 15;

+DEFAULT.baselinestart   = 0;

+DEFAULT.baselineend     = 0;

 DEFAULT.trfactor        = 0.5;

 DEFAULT.frameshiftstart = -5;

-DEFAULT.frameshiftend   = 35;

+DEFAULT.frameshiftend   = 15;

 DEFAULT.frameshiftdur   = 0;

 DEFAULT.classdefstring  = 'left,\t[9,11,13]\nright,\t[10,12,14]';

 DEFAULT.voxelstring     = 'M1 l + 3 \nM1 r + 3\n';

@@ -8,7 +8,8 @@ if(nSubjects < 2)

     error('SVMCrossVal:xsvmSubjectLoop:tooFewSubjects','You need at least 2 Subjects in this Across-Subject analysis!');

 end

-RANDOMIZE_DATAPOINTS = 1;

+RANDOMIZE_DATAPOINTS = header.svmrnd;

+NAN_AS_ZERO = header.nantozero;

 decode = struct;

 decode.decodePerformance = [];

@@ -65,6 +66,11 @@ for timeIndex = 1:timeLineEnd-timeLineStart+1

             svm_train_label  = svm_train_label(rndindex);

         end

+        if NAN_AS_ZERO

+            svm_train_data(isnan(svm_train_data))=0;

+        end

+        

+

         svmmodel = svmtrain(svm_train_label,svm_train_data,svmopts);

         [plabel accuracy dvalue] = svmpredict(svm_validation_label,svm_validation_data,svmmodel,'');