SVMCrossVal.git

SOM Single run
Parent d0ee75c
Christoph Budziszewski

Christoph Budziszewski commited on 2009-04-07 16:43:52
Zeige 4 geänderte Dateien mit 83 Einfügungen und 83 Löschungen. 

git-svn-id: https://svn.discofish.de/MATLAB/spmtoolbox/SVMCrossVal@174 83ab2cfd-5345-466c-8aeb-2b2739fb922d
private/main.m
Zeige Datei @ 916fe4f
... ... 
@@ -133,8 +133,12 @@ switch task
133 133 
         assignin('base','decode',decode);
134 134 
     case 'SOM'
135 135 
         display('SOM');
136 
-        disp('not implemented')
137 
-
136 
+        somopts.size = [3 3];
137 
+        somopts.lattice = 'rect';
138 
+        somopts.nfold = 6;
139 
+        decode = som_subject_batch(header,data,somopts);
140 
+        decode.header = header;
141 
+        assignin('base','decode',decode);
138 142 
     case 'XSOM'
139 143 
         display('XSOM');
140 144 
         somopts.size = [3 3];
private/som_combined_subject_batch.m
Zeige Datei @ d0ee75c
... ... 
@@ -1,80 +0,0 @@
1 
-%% subject loop
2 
-function decode = som_combined_subject_batch(header,subjectdata,somOpts)
3 
-
4 
-somOpts.size = [3 3];
5 
-somOpts.lattice = 'rect';
6 
-
7 
-addpath 'somtoolbox2';
8 
-
9 
-% nSubjects = numel(subjectdata);
10 
-% if(nSubjects < 2)
11 
-%     error('SVMCrossVal:xsvmSubjectLoop:tooFewSubjects','You need at least 2 Subjects in this Across-Subject analysis!');
12 
-% end
13 
-
14 
-RANDOMIZE_DATAPOINTS = 1;
15 
-
16 
-decode = struct;
17 
-decode.decodePerformance = [];
18 
-decode.rawTimeCourse     = [];
19 
-
20 
-disp(sprintf('computinig additional datastructs for %u subjects',nSubjects));
21 
-
22 
-timeline = header.timeline;
23 
-timeline.frameShiftStart = header.frameShift.frameShiftStart;
24 
-timeline.frameShiftEnd   = header.frameShift.frameShiftEnd;
25 
-timeline.decodeDuration  = header.frameShift.decodeDuration;
26 
-
27 
-% TimePointMatrix
28 
-for subjectDataID = 1:nSubjects
29 
-    currentSubject = subjectdata{subjectDataID};
30 
-    timePointArgs.pst           = currentSubject.pst;
31 
-    timePointArgs.labelMap      = LabelMap(header.classDef.labelCells,header.classDef.conditionCells);
32 
-    timePointArgs.eventList     = header.classDef.eventMatrix;
33 
-
34 
-    timePointMatrix{subjectDataID} = buildTimePointMatrix(timeline,timePointArgs);
35 
-
36 
-    decode.rawTimeCourse = [decode.rawTimeCourse currentSubject.pst];
37 
-end
38 
-
39 
-% timeframe x-subject validation
40 
-timeLineStart   = timeline.frameShiftStart;
41 
-timeLineEnd     = timeline.frameShiftEnd;
42 
-
43 
-display(sprintf('%u -fold cross validation for %u timeslices.\n',nSubjects,size(1:timeLineEnd-timeLineStart+1,2)));
44 
-% disp(sprintf('Press ANY-Key to continue.\n Use Retrun if your Keyboard lacks the ANY-Key.'));
45 
-% pause
46 
-
47 
-for timeIndex = 1:timeLineEnd-timeLineStart+1
48 
-    cross_value = [];
49 
-    for validationSubjectID = 1:nSubjects
50 
-        svm_train_label = [];
51 
-        svm_train_data  = [];
52 
-        svm_validation_label = [];
53 
-        svm_validation_data  = [];
54 
-        for subjectDataID = 1:nSubjects
55 
-            svmstruct = calculateSVMTables(timePointMatrix{subjectDataID},timeIndex);
56 
-            if subjectDataID == validationSubjectID
57 
-                svm_validation_label = svmstruct.svmlabel;
58 
-                svm_validation_data  = svmstruct.svmdata;
59 
-            else
60 
-                svm_train_label = [svm_train_label; svmstruct.svmlabel];
61 
-                svm_train_data  = [svm_train_data;  svmstruct.svmdata];
62 
-            end
63 
-        end
64 
-
65 
-        if RANDOMIZE_DATAPOINTS
66 
-            rndindex  = randperm(length(svm_train_label));
67 
-            svm_train_data   = svm_train_data(rndindex,:);
68 
-            svm_train_label  = svm_train_label(rndindex);
69 
-        end
70 
-
71 
-        [sD sM] = train_som(svm_train_label, svm_train_data, somOpts);
72 
-
73 
-        performance = som_decode(sD, sM, svm_validation_data,svm_validation_label);
74 
-
75 
-        cross_value = [cross_value performance];
76 
-    decode.decodePerformance = [decode.decodePerformance; cross_value];
77 
-end
78 
-   display('decode done');
79 
-end
80 
-
private/som_subject_batch.m
Zeige Datei @ 916fe4f
... ... 
@@ -0,0 +1,76 @@
1 
+%% subject loop
2 
+function decode = som_subject_batch(header,subjectdata,somOpts)
3 
+
4 
+addpath 'somtoolbox2';
5 
+
6 
+RANDOMIZE_DATAPOINTS = 1;
7 
+
8 
+decode = struct;
9 
+decode.decodePerformance = [];
10 
+decode.rawTimeCourse     = [];
11 
+
12 
+nSubjects = numel(subjectdata);
13 
+disp(sprintf('batch processing %u subjects',nSubjects));
14 
+
15 
+timeline = header.timeline;
16 
+timeline.frameShiftStart = header.frameShift.frameShiftStart;
17 
+timeline.frameShiftEnd   = header.frameShift.frameShiftEnd;
18 
+timeline.decodeDuration  = header.frameShift.decodeDuration;
19 
+
20 
+
21 
+timeLineStart   = timeline.frameShiftStart;
22 
+timeLineEnd     = timeline.frameShiftEnd;
23 
+
24 
+% TimePointMatrix
25 
+for subjectDataID = 1:nSubjects
26 
+    currentSubject = subjectdata{subjectDataID};
27 
+    timePointArgs.pst           = currentSubject.pst;
28 
+    timePointArgs.labelMap      = LabelMap(header.classDef.labelCells,header.classDef.conditionCells);
29 
+    timePointArgs.eventList     = header.classDef.eventMatrix;
30 
+
31 
+    timePointMatrix{subjectDataID} = buildTimePointMatrix(timeline,timePointArgs);
32 
+
33 
+    decode.rawTimeCourse = [decode.rawTimeCourse currentSubject.pst];
34 
+
35 
+
36 
+    display(sprintf('%u -fold cross validation for %u timeslices.\n',somOpts.nfold,size(1:timeLineEnd-timeLineStart+1,2)));
37 
+
38 
+
39 
+    for timeIndex = 1:timeLineEnd-timeLineStart+1
40 
+        svmstruct = calculateSVMTables(timePointMatrix{subjectDataID},timeIndex);
41 
+        nElements = length(svmstruct.svmlabel);
42 
+
43 
+        if RANDOMIZE_DATAPOINTS
44 
+            rndindex  = randperm(nElements);
45 
+            svmstruct.svmdata   = svmstruct.svmdata(rndindex,:);
46 
+            svmstruct.svmlabel  = svmstruct.svmlabel(rndindex);
47 
+        end
48 
+
49 
+        chunksize = nElements / somOpts.nfold;
50 
+
51 
+        cross_value = 0;
52 
+        for iFold = 1:somOpts.nfold
53 
+            chunkstart = (iFold-1)*chunksize+1;
54 
+            chunkend   = min(iFold*chunksize,nElements);
55 
+
56 
+            svm_train_label = svmstruct.svmlabel;
57 
+            svm_train_data  = svmstruct.svmdata;
58 
+
59 
+            svm_validation_label = svmstruct.svmlabel(chunkstart:chunkend);
60 
+            svm_validation_data  = svmstruct.svmdata(chunkstart:chunkend,:);
61 
+
62 
+            svm_train_label(chunkstart:chunkend) = []; %del test set
63 
+            svm_train_data(chunkstart:chunkend,:) = [];% del test set
64 
+
65 
+            [sD sM] = som_train(svm_train_label, svm_train_data, somOpts);
66 
+
67 
+            performance = som_decode(sM, svm_validation_data,svm_validation_label);
68 
+
69 
+            cross_value = [cross_value, performance];
70 
+        end
71 
+        decode.decodePerformance = [decode.decodePerformance; cross_value];
72 
+    end
73 
+end
74 
+display('decode done');
75 
+end
76 
+
private/ui_main.m
Zeige Datei @ 916fe4f
... ... 
@@ -346,7 +346,7 @@ pSOM = uipanel(parent,'Units','normalized','Position',[0.5 0.4 0.5 0.4]);
346 346 
     btnRunSOM = uicontrol(pSOM,'String','run SOM Crossvalidation',...
347 347 
         'Units','normalized',...
348 348 
         'Position',[0.0 0.25 1 0.25]);
349 
-    set(btnRunSOM,'Enable','off');
349 
+    set(btnRunSOM,'Enable','on');
350 350
351 351 
     btnRunXSOM = uicontrol(pSOM,'String','run SOM X-Subject validation',...
352 352 
         'Units','normalized',...
353 353