single subject -> svm classification -> plotDecodePerformance working! (d0eb71f)

single subject -> svm classification -> plotDecodePerformance working!

Christoph Budziszewski commited on 2009-03-06 13:17:25
Zeige 5 geänderte Dateien mit 25 Einfügungen und 20 Löschungen.

git-svn-id: https://svn.discofish.de/MATLAB/spmtoolbox/SVMCrossVal@145 83ab2cfd-5345-466c-8aeb-2b2739fb922d

@LabelMap/lm_getSVMLabel.m cff3c28..e27120e
private/calculateDecodePerformance.m fc98933..3a4f842
private/calculateMultiSubjectDecodePerformance.m 41145fe..2e37e9d
private/main.m cc94521..634f644
private/plotDecodePerformance.m 4c58cc3..8c03c02

@LabelMap/lm_getSVMLabel.m

Zeige Datei @ d0eb71f

...	...	@@ -1,3 +1,3 @@
1	1	function svmlabel = lm_getSVMLabel(mapping,condition)
2		- svmlabel = getValue(mapping,lm_getLabel(mapping,condition));
	2	+ svmlabel = lm_getValue(mapping,lm_getLabel(mapping,condition));
3	3	end
4	4	\ No newline at end of file

private/calculateDecodePerformance.m

Zeige Datei @ d0eb71f

@@ -1,7 +1,8 @@
-function outputStruct = calculateDecodePerformance(timeline,subjectStruct,svmopts)
+function outputStruct = calculateDecodePerformance(header,subjectStruct,svmopts)
 outputStruct = struct;
 RANDOMIZE_DATAPOINTS = 1;
 
+timeline = header.timeline;
 
 timeLineStart   = timeline.frameShiftStart;
 timeLineEnd     = timeline.frameShiftEnd;
@@ -16,8 +17,8 @@ timeLineEnd     = timeline.frameShiftEnd;
 % end
 
 timePointArgs.pst           = subjectStruct.pst;
-timePointArgs.labelMap      = labelMap;
-timePointArgs.eventList     = eventList;
+timePointArgs.labelMap      = LabelMap(header.classDef.labelCells,header.classDef.conditionCells);
+timePointArgs.eventList     = header.classDef.eventMatrix;
 
 timePointMatrix = buildTimePointMatrix(timeline,timePointArgs);
 
@@ -41,9 +42,9 @@ end
 outputStruct.decodePerformance  = decodePerformance;
 outputStruct.svmdata            = svmdata;
 outputStruct.svmlabel           = svmlabel;
-outputStruct.rawTimeCourse      = pst;
-outputStruct.minPerformance     = minPerformance;
-outputStruct.maxPerformance     = maxPerformance;
+outputStruct.rawTimeCourse      = subjectStruct.pst;
+% outputStruct.minPerformance     = minPerformance;
+% outputStruct.maxPerformance     = maxPerformance;
 end
 
 

private/calculateMultiSubjectDecodePerformance.m

Zeige Datei @ d0eb71f

@@ -18,7 +18,7 @@ for subjectDataID = 1:size(subjectdata)
     warning_state               = warning('off','all');
     display('calculating ...');
     
-        decode.(namehelper)         = calculateDecodePerformance(header.timeline,currentSubject,svmopts);
+        decode.(namehelper)         = calculateDecodePerformance(header,currentSubject,svmopts);
 
     display('... done');
     display('restoring warnings');

private/main.m

Zeige Datei @ d0eb71f

@@ -71,7 +71,7 @@ switch task
         disp('SVM');
         svmopts    = getSvmArgs(model,1);
         preprocessedData = evalin('base','preprocessedData');
-        calculateMultiSubjectDecodePerformance(preprocessedData.header,preprocessedData.subjectdata,svmopts);
+        decode = calculateMultiSubjectDecodePerformance(preprocessedData.header,preprocessedData.subjectdata,svmopts);
         
         
     case 'X-SVM'

private/plotDecodePerformance.m

Zeige Datei @ d0eb71f

@@ -1,21 +1,25 @@
-function plotDecodePerformance(timeline,inputStruct)
+function plotDecodePerformance(header,decode,subjectData)
 
-global CROSSVAL_METHOD_DEF;
+global SVMCROSSVAL_CROSSVAL_METHOD_DEF;
 
 PSTH_AXIS_MIN = -1;
 PSTH_AXIS_MAX = 1;
 
+timeline = header.timeline;
+
         psthStart         = timeline.psthStart;
         psthEnd           = timeline.psthEnd;
         frameStart        = timeline.frameShiftStart;
         frameEnd          = timeline.frameShiftEnd;
 
-        nClasses          = inputStruct.nClasses;
-        decodePerformance = inputStruct.decodePerformance;
-        psth              = inputStruct.rawTimeCourse;
-        SubjectID         = inputStruct.SubjectID;
-        smoothed          = inputStruct.smoothed;
-        PLOT_METHOD       = inputStruct.CROSSVAL_METHOD;
+        nClasses          = numel(header.classDef.labelCells);
+        decodePerformance = decode.decodePerformance;
+        psth              = decode.rawTimeCourse;
+        SubjectID         = subjectData;
+
+        smoothed          = 'yes';
+        
+        PLOT_METHOD       = SVMCROSSVAL_CROSSVAL_METHOD_DEF.svmcrossval;
 %         CROSSVAL_METHOD_DEF = inputStruct.CROSSVAL_METHOD_DEF;
 
 
@@ -50,13 +54,13 @@ PSTH_AXIS_MAX = 1;
     
     
     switch PLOT_METHOD
-        case CROSSVAL_METHOD_DEF.svmcrossval
+        case SVMCROSSVAL_CROSSVAL_METHOD_DEF.svmcrossval
             plot(frameStart:frameEnd, mean(decodePerformance,2) ,'b','LineWidth',2);
 
             se = myStdErr(decodePerformance,2);
             plot(frameStart:frameEnd, mean(decodePerformance,2)+se ,'b:');
             plot(frameStart:frameEnd, mean(decodePerformance,2)-se ,'b:');
-        case CROSSVAL_METHOD_DEF.classPerformance
+        case SVMCROSSVAL_CROSSVAL_METHOD_DEF.classPerformance
             for c = 1:size(decodePerformance,2)
                 plot(frameStart:frameEnd, decodePerformance(:,c) ,[colorChooser(mod(c,nClasses)+3) '-']);
             end
@@ -78,7 +82,7 @@ PSTH_AXIS_MAX = 1;
     end
     
     if nSubjects == 1
-        subjectName = cell2Mat(SubjectID);
+        subjectName = SubjectID{1}.name;
         title = sprintf('Subject %s, over %g voxel, %s',subjectName,nVoxelPerSubject,smoothedString);
     else
         title = sprintf('%g Subjects, %g Voxel per Subject, %s',nSubjects,nVoxelPerSubject,smoothedString);


...	...	@@ -1,7 +1,8 @@
1		-function outputStruct = calculateDecodePerformance(timeline,subjectStruct,svmopts)
	1	+function outputStruct = calculateDecodePerformance(header,subjectStruct,svmopts)
2	2	outputStruct = struct;
3	3	RANDOMIZE_DATAPOINTS = 1;
4	4
	5	+timeline = header.timeline;
5	6
6	7	timeLineStart = timeline.frameShiftStart;
7	8	timeLineEnd = timeline.frameShiftEnd;
...	...	@@ -16,8 +17,8 @@ timeLineEnd = timeline.frameShiftEnd;
16	17	% end
17	18
18	19	timePointArgs.pst = subjectStruct.pst;
19		-timePointArgs.labelMap = labelMap;
20		-timePointArgs.eventList = eventList;
	20	+timePointArgs.labelMap = LabelMap(header.classDef.labelCells,header.classDef.conditionCells);
	21	+timePointArgs.eventList = header.classDef.eventMatrix;
21	22
22	23	timePointMatrix = buildTimePointMatrix(timeline,timePointArgs);
23	24
...	...	@@ -41,9 +42,9 @@ end
41	42	outputStruct.decodePerformance = decodePerformance;
42	43	outputStruct.svmdata = svmdata;
43	44	outputStruct.svmlabel = svmlabel;
44		-outputStruct.rawTimeCourse = pst;
45		-outputStruct.minPerformance = minPerformance;
46		-outputStruct.maxPerformance = maxPerformance;
	45	+outputStruct.rawTimeCourse = subjectStruct.pst;
	46	+% outputStruct.minPerformance = minPerformance;
	47	+% outputStruct.maxPerformance = maxPerformance;
47	48	end
48	49
49	50

...	...	@@ -18,7 +18,7 @@ for subjectDataID = 1:size(subjectdata)
18	18	warning_state = warning('off','all');
19	19	display('calculating ...');
20	20
21		- decode.(namehelper) = calculateDecodePerformance(header.timeline,currentSubject,svmopts);
	21	+ decode.(namehelper) = calculateDecodePerformance(header,currentSubject,svmopts);
22	22
23	23	display('... done');
24	24	display('restoring warnings');

...	...	@@ -71,7 +71,7 @@ switch task
71	71	disp('SVM');
72	72	svmopts = getSvmArgs(model,1);
73	73	preprocessedData = evalin('base','preprocessedData');
74		- calculateMultiSubjectDecodePerformance(preprocessedData.header,preprocessedData.subjectdata,svmopts);
	74	+ decode = calculateMultiSubjectDecodePerformance(preprocessedData.header,preprocessedData.subjectdata,svmopts);
75	75
76	76
77	77	case 'X-SVM'

...	...	@@ -1,21 +1,25 @@
1		-function plotDecodePerformance(timeline,inputStruct)
	1	+function plotDecodePerformance(header,decode,subjectData)
2	2
3		-global CROSSVAL_METHOD_DEF;
	3	+global SVMCROSSVAL_CROSSVAL_METHOD_DEF;
4	4
5	5	PSTH_AXIS_MIN = -1;
6	6	PSTH_AXIS_MAX = 1;
7	7
	8	+timeline = header.timeline;
	9	+
8	10	psthStart = timeline.psthStart;
9	11	psthEnd = timeline.psthEnd;
10	12	frameStart = timeline.frameShiftStart;
11	13	frameEnd = timeline.frameShiftEnd;
12	14
13		- nClasses = inputStruct.nClasses;
14		- decodePerformance = inputStruct.decodePerformance;
15		- psth = inputStruct.rawTimeCourse;
16		- SubjectID = inputStruct.SubjectID;
17		- smoothed = inputStruct.smoothed;
18		- PLOT_METHOD = inputStruct.CROSSVAL_METHOD;
	15	+ nClasses = numel(header.classDef.labelCells);
	16	+ decodePerformance = decode.decodePerformance;
	17	+ psth = decode.rawTimeCourse;
	18	+ SubjectID = subjectData;
	19	+
	20	+ smoothed = 'yes';
	21	+
	22	+ PLOT_METHOD = SVMCROSSVAL_CROSSVAL_METHOD_DEF.svmcrossval;
19	23	% CROSSVAL_METHOD_DEF = inputStruct.CROSSVAL_METHOD_DEF;
20	24
21	25
...	...	@@ -50,13 +54,13 @@ PSTH_AXIS_MAX = 1;
50	54
51	55
52	56	switch PLOT_METHOD
53		- case CROSSVAL_METHOD_DEF.svmcrossval
	57	+ case SVMCROSSVAL_CROSSVAL_METHOD_DEF.svmcrossval
54	58	plot(frameStart:frameEnd, mean(decodePerformance,2) ,'b','LineWidth',2);
55	59
56	60	se = myStdErr(decodePerformance,2);
57	61	plot(frameStart:frameEnd, mean(decodePerformance,2)+se ,'b:');
58	62	plot(frameStart:frameEnd, mean(decodePerformance,2)-se ,'b:');
59		- case CROSSVAL_METHOD_DEF.classPerformance
	63	+ case SVMCROSSVAL_CROSSVAL_METHOD_DEF.classPerformance
60	64	for c = 1:size(decodePerformance,2)
61	65	plot(frameStart:frameEnd, decodePerformance(:,c) ,[colorChooser(mod(c,nClasses)+3) '-']);
62	66	end
...	...	@@ -78,7 +82,7 @@ PSTH_AXIS_MAX = 1;
78	82	end
79	83
80	84	if nSubjects == 1
81		- subjectName = cell2Mat(SubjectID);
	85	+ subjectName = SubjectID{1}.name;
82	86	title = sprintf('Subject %s, over %g voxel, %s',subjectName,nVoxelPerSubject,smoothedString);
83	87	else
84	88	title = sprintf('%g Subjects, %g Voxel per Subject, %s',nSubjects,nVoxelPerSubject,smoothedString);
85	89

SVMCrossVal.git