|
|
|
|
|
clear all; clc
|
|
|
|
|
|
|
|
|
RandStream.setGlobalStream(RandStream('mt19937ar','seed',sum(100*clock)));
|
|
|
|
|
|
|
|
|
cd('Y:\EEG_Data\CLASSIFY\Classify for Arun\');
|
|
|
|
|
|
% Load up IDs, match with CTL IDs, get correlates
|
|
|
[EEG_IDs,~,~]=xlsread('All_POWER_ROI_INFO.xlsx','Subjects_IDs'); % Col 1 = Pd, Col 2 = CTL
|
|
|
[VAR_DATA,VAR_HDR,VAR_BOTH]=xlsread('PD_CONFLICT_VARS.xlsx');
|
|
|
% Match up
|
|
|
PD_ID=VAR_DATA(:,1);
|
|
|
MATCH_ID=VAR_DATA(:,5);
|
|
|
YrsDx=VAR_DATA(:,9);
|
|
|
if sum( EEG_IDs(:,1)==PD_ID )==28 % If PD are numerically aligned in both data sets
|
|
|
for matchi=1:28
|
|
|
MATCHTOEEG(matchi)=find( EEG_IDs(:,2)==MATCH_ID(matchi) ); % Find the ctl subj that matches with this patient
|
|
|
end
|
|
|
end
|
|
|
|
|
|
% Load EEG Data
|
|
|
SHEET={'Cue_locked_Conf','Response_locked_Conf','Response-locked_CorrectError','PostCorrectError','Alpha','RelativeAlpha'};
|
|
|
for sheeti=1:6
|
|
|
[DATA{sheeti},HDR{sheeti},BOTH{sheeti}]=xlsread('All_POWER_ROI_INFO.xlsx',SHEET{sheeti});
|
|
|
end
|
|
|
|
|
|
% ********* Setup Data ********* Do you want control only, or control-benign condi diffs?
|
|
|
for sheeti=1:4
|
|
|
CTL(:,sheeti)=DATA{sheeti}(:,3) % -DATA{sheeti}(:,2);
|
|
|
ON(:,sheeti)=DATA{sheeti}(:,5) % -DATA{sheeti}(:,4);
|
|
|
OFF(:,sheeti)=DATA{sheeti}(:,7) % -DATA{sheeti}(:,6);
|
|
|
end
|
|
|
% Alpha
|
|
|
for sheeti=5:6
|
|
|
CTL(:,sheeti)=DATA{sheeti}(:,1) ;
|
|
|
ON(:,sheeti)=DATA{sheeti}(:,2) ;
|
|
|
OFF(:,sheeti)=DATA{sheeti}(:,3) ;
|
|
|
end
|
|
|
|
|
|
|
|
|
% ********* Setup Contrasts ********* InData should have Col 1 = group (1=patient, 0=Ctl) and Cols 2-N are data
|
|
|
InData=[ [ones(28,1),CTL] ; [zeros(28,1),ON] ] ;
|
|
|
TITLE='CTL_ON_NoDiff';
|
|
|
iterations=500;
|
|
|
vars=[2,3,4,5,6] % [2,3,4,5,7]; % Column 1 is group, 2-N are variables to select any number of here
|
|
|
% ********* ***** *********
|
|
|
|
|
|
% Classify
|
|
|
for Xvali=1:3
|
|
|
if Xvali==1, Xval='5X';
|
|
|
elseif Xvali==2, Xval='10X';
|
|
|
elseif Xvali==3, Xval='LOO';
|
|
|
end
|
|
|
Classify_Scalars_SVM(InData,TITLE,iterations,vars,Xval);
|
|
|
Classify_Scalars_LASSO(InData,TITLE,iterations,vars,Xval); % not fully validated, just playing with this for now.
|
|
|
end
|
|
|
% Now each paired with their best match CTL
|
|
|
Classify_Scalars_SVM_MatchSubjs(InData,TITLE,MATCHTOEEG,vars,'Match');
|
|
|
Classify_Scalars_LASSO_MatchSubjs(InData,TITLE,MATCHTOEEG,vars,'Match');
|
|
|
|
|
|
%% Aggregate Different Predictors / SVM
|
|
|
|
|
|
% ########################
|
|
|
Xval='LOO';
|
|
|
iterations=500; % If 'Match', iterations needs to = 28
|
|
|
Classifier='SVM'; % 'SVM' 'LASSO'
|
|
|
% TITLE={'CTL_ON_Cue','CTL_ON_Resp','CTL_ON_Err','CTL_ON_PE','CTL_ON_RelAlpha','CTL_ON'};
|
|
|
TITLE={'CTL_OFF_Cue','CTL_OFF_Resp','CTL_OFF_Err','CTL_OFF_PE','CTL_OFF_RelAlpha','CTL_OFF'};
|
|
|
% TITLE={'CTL_ON_NoDiff_Cue','CTL_ON_NoDiff_Resp','CTL_ON_NoDiff_Err','CTL_ON_NoDiff_PE','CTL_ON_NoDiff_Alpha','CTL_ON_NoDiff'}; % not C-I diff
|
|
|
% ########################
|
|
|
|
|
|
for vari=1:6
|
|
|
|
|
|
if strmatch(Classifier,'SVM')
|
|
|
load(['SVM_',TITLE{vari},'_',Xval,'_iter',num2str(iterations),'.mat']);
|
|
|
OUTPUTS(vari,1)=mean(mean(Aset_acc));
|
|
|
OUTPUTS(vari,2)=mean(mean(Bset_acc));
|
|
|
if strmatch(Xval,'Match')
|
|
|
SCORES(vari,1,:)=Aset_score;
|
|
|
SCORES(vari,2,:)=Bset_score;
|
|
|
else
|
|
|
SCORES(vari,1,:)=nanmean(Aset_score');
|
|
|
SCORES(vari,2,:)=nanmean(Bset_score');
|
|
|
end
|
|
|
clear A* B*; classifier=1;
|
|
|
|
|
|
elseif strmatch(Classifier,'LASSO')
|
|
|
load(['LASSO_',TITLE{vari},'_',Xval,'_iter',num2str(iterations),'.mat']);
|
|
|
OUTPUTS(vari,1)=mean(LASSO_Probability_Tst2(:,2));
|
|
|
OUTPUTS(vari,2)=mean(LASSO_Probability_Tst2(:,3));
|
|
|
if strmatch(Xval,'Match')
|
|
|
if vari==6
|
|
|
SCORES(vari,1,:)=LASSO_Betas(LASSO_Betas==max(abs(LASSO_Betas)')');
|
|
|
SCORES(vari,2,:)=LASSO_Betas(LASSO_Betas==max(abs(LASSO_Betas)')');
|
|
|
else
|
|
|
SCORES(vari,1,:)=LASSO_Betas;
|
|
|
SCORES(vari,2,:)=LASSO_Betas;
|
|
|
end
|
|
|
end
|
|
|
clear A* B*; classifier=2;
|
|
|
end
|
|
|
|
|
|
end
|
|
|
|
|
|
|
|
|
% Plot
|
|
|
COL={'m','y'};
|
|
|
SHAPE={'d','o','s','p','h','^'};
|
|
|
|
|
|
figure;
|
|
|
subplot(1,3,1:2);
|
|
|
hold on
|
|
|
for vari=1:6
|
|
|
plot(1-OUTPUTS(vari,2),OUTPUTS(vari,1),...
|
|
|
SHAPE{vari},'MarkerFaceColor','k','MarkerEdgeColor',COL{classifier},'MarkerSize',10)
|
|
|
end
|
|
|
set(gca,'ylim',[0 1],'ytick',[0:.1:1],'xlim',[0 1],'xtick',[0:.1:1]);
|
|
|
legend(TITLE,'location','southeast', 'Interpreter', 'none');
|
|
|
ylabel('Sensitivity'); xlabel('1-Specificity');
|
|
|
plot([0 1],[0 1],'k');
|
|
|
title(['Classification of ',TITLE{6}], 'Interpreter', 'none')
|
|
|
|
|
|
subplot(1,3,3);
|
|
|
hold on
|
|
|
for vari=1:6
|
|
|
bar(vari,mean(OUTPUTS(vari,:),2),.4,COL{classifier})
|
|
|
end
|
|
|
set(gca,'ylim',[.4 1],'ytick',[.4:.1:1],'xlim',[0 7],'xtick',[1:1:6],'xticklabels',TITLE);
|
|
|
xtickangle(90)
|
|
|
title('Average')
|
|
|
|
|
|
if strmatch(Classifier,'SVM')
|
|
|
figure;
|
|
|
for paneli=1:6
|
|
|
subplot(2,3,paneli); hold on
|
|
|
scatter(YrsDx,abs(squeeze(SCORES(paneli,1,:)))); lsline
|
|
|
[rho,p]=corr(abs(squeeze(SCORES(paneli,1,:))),YrsDx,'type','Spearman');
|
|
|
text(.1,.1,['rho=',num2str(rho),' p=',num2str(p)],'sc');
|
|
|
xlabel('YrsDx'); ylabel('confidence');
|
|
|
title(TITLE{paneli}, 'Interpreter', 'none')
|
|
|
end
|
|
|
end
|
|
|
|
|
|
% Aggregate all X-Vals and Algorithms
|
|
|
ToAgg=6; iterations=500;
|
|
|
|
|
|
for Xvali=1:3
|
|
|
if Xvali==1, Xval='5X';
|
|
|
elseif Xvali==2, Xval='10X';
|
|
|
elseif Xvali==3, Xval='LOO';
|
|
|
end
|
|
|
load(['SVM_',TITLE{ToAgg},'_',Xval,'_iter',num2str(iterations),'.mat']);
|
|
|
OUTPUT_AGG(1,Xvali,1)=mean(mean(Aset_acc));
|
|
|
OUTPUT_AGG(1,Xvali,2)=mean(mean(Bset_acc));
|
|
|
clear A* B*;
|
|
|
|
|
|
load(['LASSO_',TITLE{ToAgg},'_',Xval,'_iter',num2str(iterations),'.mat']);
|
|
|
OUTPUT_AGG(2,Xvali,1)=mean(LASSO_Probability_Tst2(:,2));
|
|
|
OUTPUT_AGG(2,Xvali,2)=mean(LASSO_Probability_Tst2(:,3));
|
|
|
clear LASSO*;
|
|
|
end
|
|
|
|
|
|
load(['SVM_',TITLE{ToAgg},'_Match_iter28.mat']);
|
|
|
OUTPUT_AGG(1,4,1)=mean(mean(Aset_acc));
|
|
|
OUTPUT_AGG(1,4,2)=mean(mean(Bset_acc));
|
|
|
clear A* B*;
|
|
|
|
|
|
load(['LASSO_',TITLE{ToAgg},'_Match_iter28.mat']);
|
|
|
OUTPUT_AGG(2,4,1)=mean(LASSO_Probability_Tst2(:,2));
|
|
|
OUTPUT_AGG(2,4,2)=mean(LASSO_Probability_Tst2(:,3));
|
|
|
clear LASSO*;
|
|
|
|
|
|
% Plot
|
|
|
COL={'m','y'};
|
|
|
SHAPE={'d','o','s','p'};
|
|
|
SHIFT=[-.2,-.1,.1,.2];
|
|
|
|
|
|
figure;
|
|
|
subplot(1,3,1:2);
|
|
|
hold on
|
|
|
for classifier=1:2
|
|
|
for Xvali=1:4
|
|
|
plot(1-OUTPUT_AGG(classifier,Xvali,2),OUTPUT_AGG(classifier,Xvali,1),...
|
|
|
SHAPE{Xvali},'MarkerFaceColor','k','MarkerEdgeColor',COL{classifier},'MarkerSize',10)
|
|
|
end
|
|
|
end
|
|
|
set(gca,'ylim',[0 1],'ytick',[0:.1:1],'xlim',[0 1],'xtick',[0:.1:1]);
|
|
|
legend({'SVM 5X','SVM 10X','SVM LOO','SVM Match','LASSO 5X','LASSO 10X','LASSO LOO','LASSO Match'},'location','southeast');
|
|
|
ylabel('Sensitivity'); xlabel('1-Specificity');
|
|
|
plot([0 1],[0 1],'k');
|
|
|
title(['Classification of ',TITLE{ToAgg}], 'Interpreter', 'none')
|
|
|
|
|
|
subplot(1,3,3);
|
|
|
hold on
|
|
|
for classifier=1:2
|
|
|
for Xvali=1:4
|
|
|
bar(classifier+SHIFT(Xvali),mean(OUTPUT_AGG(classifier,Xvali,:),3),.4,COL{classifier})
|
|
|
end
|
|
|
end
|
|
|
set(gca,'ylim',[.5 1],'ytick',[.5:.1:1],'xlim',[0 3],'xtick',[1:1:2],'xticklabels',{'SVM','LASSO'});
|
|
|
title('Average')
|
|
|
|
|
|
|
|
|
%%
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
