online.m

function [Trials,SmoothedTotal]=online(Session,Mu,Sigma,classifier,FeaturesIndeces,alpha)
%{
Inputs: whole signal 4th run (s{1,4}=Session), mu, sigma, classifier, FeaturesIndeces


%}
global SubjectID

f=4:2:40;

%%Divide the run into the (30) trials
TrialStart=Session.event.position(Session.event.name==400);%400=MI initiation
TrialEnd=Session.event.position(Session.event.name==700);%700=Relax phase

Trials=[];
for i=1:(length(TrialStart))
    Trials{i}=Session.data(TrialStart(i):TrialEnd(i),:)';
end



%%Divide each trial into 1-second windows shifted by 0.0625s
%(1 second=512 points; 0.0625 s = 32 points)
Aux=Trials;
Trials=[];
for i=1:length(Aux)
j = 1;
    for t = 512:32:(size(Aux{i},2)-mod(size(Aux{i},2),32)) %we remove the last window (not of 1 second)
        for idxChannels = 1:16
            Trials{i}.Windows(idxChannels,:,j)=Aux{i}(idxChannels,t-512+1:t);
        end 
        j= j+1;
    end
end

%% Real pseudoonline
Evidence=[];
for i=1:length(Trials)
   Evidence{i}=zeros(1,size(Trials{i}.Windows,3)+1);
   Evidence{1,i}(1)=0.5; %for each trial we start by 0.5
   Trials{i}.Features=zeros(size(Trials{i}.Windows,3),304);
   
   for j=1:size(Trials{i}.Windows,3)
       %%CAR filtering
       mean_channels=[];
       mean_channels = mean(Trials{i}.Windows(:,:,j));
       Trials{i}.Windows(:,:,j) = Trials{i}.Windows(:,:,j)- mean_channels;
       
       %%pwelch (already reshaped)
       for idxChannels=1:16
            %[psd,freqgrid] = pwelch(Trials{i}.Windows(idxChannels,:,j),0.5*512,0.4375*512,[],512);
            %psd=log(psd);
            %[freqs, idfreqs] = intersect(freqgrid,f);
            %psd = psd(idfreqs);
            %Trials{i}.Features(j,(19*(idxChannels-1)+1):19*idxChannels)=psd;
            Trials{i}.Features(j,(19*(idxChannels-1)+1):19*idxChannels)=pwelch(Trials{i}.Windows(idxChannels,:,j),0.5*512,0.25*512,f,512);%0.4375*512
       end
       for idxChannels=1:16
            Trials{i}.Features(j,19*(idxChannels-1)+1:19*idxChannels) = log(Trials{i}.Features(j,19*(idxChannels-1)+1:19*idxChannels));
       end 
       
       %%normalization
       Trials{i}.Features(j,:)=(Trials{i}.Features(j,:)-Mu)./Sigma;
       
       %%predict (posterior prob)
       [Trials{i}.Predicted(j),Trials{i}.PosteriorProb(j,:),~]=predict(classifier,Trials{i}.Features(j,FeaturesIndeces));

       
       %%smoothing
       Evidence{1,i}(j+1)=Evidence{1,i}(j)*alpha+(1-alpha)*Trials{i}.PosteriorProb(j,2);
       fprintf('Trial %d Window %d \n',i,j);

   end
   
end

%% Concatenate all the smoothed probabilities
SmoothedTotal=[];
for i=1:length(Trials)
   SmoothedTotal=cat(2,SmoothedTotal,Evidence{1,i}); 
end

figure
sz = 4;
c = [0,0,1];
scatter((1-0.5*length(Evidence{1,1})*30/length(SmoothedTotal)):(30/length(SmoothedTotal)):(31-30/length(SmoothedTotal)-0.5*length(Evidence{1,1})*30/length(SmoothedTotal)),SmoothedTotal,sz,c,'filled')
hold on
cont=1-0.5*length(Evidence{1,1})*30/length(SmoothedTotal);
for i=1:length(Trials)
    if cont~=1-0.5*length(Evidence{1,1})*30/length(SmoothedTotal)
        vline(cont,'r','');
%     else
%         vline(cont,'k','');
    end
    cont=cont+length(Evidence{1,i})*30/length(SmoothedTotal);
end
xlabel('Trials')
ylabel('Smoothed probability')
title(sprintf('Accumulated Evidence -  %s', SubjectID))
xlim([1-0.5*length(Evidence{1,1})*30/length(SmoothedTotal) 31-30/length(SmoothedTotal)-0.5*length(Evidence{1,1})*30/length(SmoothedTotal)])
%set(gca,'TickLength',[0 0]);
set(gca,'xtick',5:5:30);

%% Examples of plots
if SubjectID == "ak6"
    
    figure
    plot([0:(length(Evidence{1,10})-1)]/16,Evidence{1,10},'b')%otherwise 6 or 7
    xlabel('Time [s]')
    ylabel('Accumulated Evidence')
    set(gca,'YLim',[0 1])
    title('Good Trial')
    
    figure
    plot([0:(length(Evidence{1,22})-1)]/16,Evidence{1,22},'b')
    xlabel('Time [s]')
    ylabel('Accumulated Evidence')
    set(gca,'YLim',[0 1])
    title('Bad Trial')
    
end



end