EmotivSMILE.m

classdef EmotivSMILE < handle
    %----------
    %EmotivSMILE
    %----------
    %queries the EDK library in order to recieve data from the
    %Emotiv EEG headset
    %
    %% Usage;
    % To create the object
    % a = EmotivEEG;
    % % To start a timer (keeps running until stopped) and get latest data
    % a.Run
    % % To plot the latest /c data variable values
    % a.Plot
    % % To stop the timmer
    % a.Stop
    % % To record 10 seconds of data and save to a file
    % a.Record(10)    
    % % To load recorded data from file into /c data variable (NOTE if you have a timer on this will be overwritten)
    % LoadRecordedData(self,filename)    
    % % To disconnect and unload the library
    % delete(a)
    

    properties (Access = public)
        sampFreq = 128;                   
        recordSeconds = 1;        
        acqisitionSeconds = 1;
        timerPeriod = 0.35; %seconds *MUST BE LESS THAN acqisitionSeconds. Also used for updating the record data
        plotTimerPeriod = 0.5;
        fig_h;
    end
    
    properties (Constant)
        unitIdentifier = 'Emotiv Systems-5'        
    end
    
    properties (SetAccess = protected)
        runAsyncTimer
        plotTimer
        channelCount
        data = [];
        eEvent
        dataRowCount
        hData % handles (pointer) to the data stored in library
        DataChannels
        EE_DataChannels_enum = struct('ED_COUNTER',0,'ED_INTERPOLATED',1,'ED_RAW_CQ',2,'ED_AF3',3,'ED_F7',4,'ED_F3',5,'ED_FC5',6,'ED_T7',7,'ED_P7',8,'ED_O1',9,'ED_O2',10,'ED_P8',11,'ED_T8',12,'ED_FC6',13,'ED_F4',14,'ED_F8',15,'ED_AF4',16,'ED_GYROX',17,'ED_GYROY',18,'ED_TIMESTAMP',19,'ED_ES_TIMESTAMP',20,'ED_FUNC_ID',21,'ED_FUNC_VALUE',22,'ED_MARKER',23,'ED_SYNC_SIGNAL',24);
%         EE_DataChannels_enum = struct('ED_COUNTER',0,'ED_AF3',1,'ED_F7',2,'ED_F3',3,'ED_FC5',4,'ED_FC6',5,'ED_F4',6,'ED_F8',7,'ED_AF4',8,'ED_TIMESTAMP',9,'ED_ES_TIMESTAMP',10);
        EE_CognitivTrainingControl_enum = struct('COG_NONE',0,'COG_START',1,'COG_ACCEPT',2,'COG_REJECT',3,'COG_ERASE',4,'COG_RESET',5);
        EE_ExpressivAlgo_enum = struct('EXP_NEUTRAL',1,'EXP_BLINK',2,'EXP_WINK_LEFT',4,'EXP_WINK_RIGHT',8,'EXP_HORIEYE',16,'EXP_EYEBROW',32,'EXP_FURROW',64,'EXP_SMILE',128,'EXP_CLENCH',256,'EXP_LAUGH',512,'EXP_SMIRK_LEFT',1024,'EXP_SMIRK_RIGHT',2048);
        EE_ExpressivTrainingControl_enum = struct('EXP_NONE',0,'EXP_START',1,'EXP_ACCEPT',2,'EXP_REJECT',3,'EXP_ERASE',4,'EXP_RESET',5);
        EE_ExpressivThreshold_enum = struct('EXP_SENSITIVITY',0);
        EE_CognitivEvent_enum = struct('EE_CognitivNoEvent',0,'EE_CognitivTrainingStarted',1,'EE_CognitivTrainingSucceeded',2,'EE_CognitivTrainingFailed',3,'EE_CognitivTrainingCompleted',4,'EE_CognitivTrainingDataErased',5,'EE_CognitivTrainingRejected',6,'EE_CognitivTrainingReset',7,'EE_CognitivAutoSamplingNeutralCompleted',8,'EE_CognitivSignatureUpdated',9);
        EE_EmotivSuite_enum=struct('EE_EXPRESSIV',0,'EE_AFFECTIV',1,'EE_COGNITIV',2);
        EE_ExpressivEvent_enum=struct('EE_ExpressivNoEvent',0,'EE_ExpressivTrainingStarted',1,'EE_ExpressivTrainingSucceeded',2,'EE_ExpressivTrainingFailed',3,'EE_ExpressivTrainingCompleted',4,'EE_ExpressivTrainingDataErased',5,'EE_ExpressivTrainingRejected',6,'EE_ExpressivTrainingReset',7);
        EE_CognitivAction_enum=struct('COG_NEUTRAL',1,'COG_PUSH',2,'COG_PULL',4,'COG_LIFT',8,'COG_DROP',16,'COG_LEFT',32,'COG_RIGHT',64,'COG_ROTATE_LEFT',128,'COG_ROTATE_RIGHT',256,'COG_ROTATE_CLOCKWISE',512,'COG_ROTATE_COUNTER_CLOCKWISE',1024,'COG_ROTATE_FORWARDS',2048,'COG_ROTATE_REVERSE',4096,'COG_DISAPPEAR',8192);
        EE_InputChannels_enum=struct('EE_CHAN_CMS',0,'EE_CHAN_DRL',1,'EE_CHAN_FP1',2,'EE_CHAN_AF3',3,'EE_CHAN_F7',4,'EE_CHAN_F3',5,'EE_CHAN_FC5',6,'EE_CHAN_T7',7,'EE_CHAN_P7',8,'EE_CHAN_O1',9,'EE_CHAN_O2',10,'EE_CHAN_P8',11,'EE_CHAN_T8',12,'EE_CHAN_FC6',13,'EE_CHAN_F4',14,'EE_CHAN_F8',15,'EE_CHAN_AF4',16,'EE_CHAN_FP2',17);
        EE_ExpressivSignature_enum=struct('EXP_SIG_UNIVERSAL',0,'EXP_SIG_TRAINED',1);
        EE_Event_enum=struct('EE_UnknownEvent',0,'EE_EmulatorError',1,'EE_ReservedEvent',2,'EE_UserAdded',16,'EE_UserRemoved',32,'EE_EmoStateUpdated',64,'EE_ProfileEvent',128,'EE_CognitivEvent',256,'EE_ExpressivEvent',512,'EE_InternalStateChanged',1024,'EE_AllEvent',2032);
        DataChannelsNames = {'ED_COUNTER','ED_INTERPOLATED','ED_RAW_CQ','ED_AF3','ED_F7','ED_F3','ED_FC5','ED_T7','ED_P7','ED_O1','ED_O2','ED_P8','ED_T8','ED_FC6','ED_F4','ED_F8','ED_AF4','ED_GYROX','ED_GYROY','ED_TIMESTAMP','ED_ES_TIMESTAMP','ED_FUNC_ID','ED_FUNC_VALUE','ED_MARKER','ED_SYNC_SIGNAL'};
%         DataChannelsNames = {'ED_COUNTER','ED_AF3','ED_F7','ED_F3','ED_FC5','ED_FC6','ED_F4','ED_F8','ED_AF4','ED_TIMESTAMP','ED_ES_TIMESTAMP'};
    end
    
    properties (SetAccess = private)
        tempData;               % holds the data while it is being read from the library
        showDebug;              % makes more show up when taking data
        ctree;                  % Classification tree
        demoResults = java.util.ArrayList();    % List of predictions during the demo mode
        responses = java.util.ArrayList();      % List of responses from machine learning
        classifyData = java.util.HashMap();     % Data for machine learning mapped to the frequency range taken at
        bandSize = 1;
        audio; 
        emotionFreqs = [167, 193, 157, 208, 244, 293, 294, 335, 388, 415,... 
            418, 135, 151, 159, 161, 163, 181, 183, 189, 150, 170, 199,...
            226, 234, 239, 245, 255, 260, 262, 268, 270, 273, 280, 281,...
            286, 289, 291, 293, 299, 304, 318, 323, 325, 331, 336, 348,...
            349, 352, 359, 362, 370, 379, 381, 405, 408]; % Based off of analysis.m with setup1 (ross) 
    end
    
    methods
%% Constructor
        function self = EmotivSMILE()
            % Check to see if library was already loaded
            if ~libisloaded('edk')    
                [notfound,warnings] = loadlibrary('edk.dll','edk.h');                 %#ok<NASGU,ASGLU>
                disp('EDK library loaded');
            else
                disp('EDK library already loaded');
            end
            
             % Success if value returned is 0
            if calllib('edk','EE_EngineConnect',int8([self.unitIdentifier 0]))
                error(['An error occured when using EE_EngineConnect to connect to ',self.unitIdentifier])
            else
                disp(['Successfully connected to ',self.unitIdentifier])
            end

            % Create a data pointer
            self.hData = calllib('edk','EE_DataCreate');
            
            % Set the record time buffer size
            calllib('edk','EE_DataSetBufferSizeInSec',self.acqisitionSeconds);
            
            % Create an events engine
            self.eEvent = calllib('edk','EE_EmoEngineEventCreate');
            
            % Set other fields
            self.dataRowCount = self.acqisitionSeconds * self.sampFreq;
            self.data = zeros(self.dataRowCount,self.channelCount);
            self.tempData = self.data; % required to work on the data variable but still allow external access to data
            self.showDebug = false;
            self.DataChannels = self.EE_DataChannels_enum; % THIS MAY BE REDUNDANT
            self.channelCount = length(self.DataChannelsNames);
            
            % Import audio files
            self.audio = struct();
            filenames = {'Detecting Happiness.wav', 'Detecting Indifference.wav', 'Detecting Sadness.wav',...
                'Happy Sad or Neither.wav', 'Would you agree.wav'};
            fields = {'positive', 'indifferent', 'negative', 'question', 'verify'};
            for i = 1:numel(fields)
                [sound, samplingRate] = audioread(filenames{i});
                self.audio.(fields{i}) = audioplayer(sound, samplingRate);
            end
        end
        
        
%% delete        
        function delete(self)
            self.Stop()
            self.StopPlot()
            
            if calllib('edk','EE_EngineDisconnect')
                warning(['Some error disconnecting from ',self.unitIdentifier]) %#ok<WNTAG>
            else
                disp(['Successfully disconnected from ',self.unitIdentifier])
            end            
            pause(0.1);
            unloadlibrary('edk');            
        end

        
%% Run
% Start a timer to continuously update the state of the data variable every /c imerPeriod
        function Run(self)
            self.runAsyncTimer = timer('TimerFcn', @(src,event)UpdateData(self), 'name', 'RunAsync','Period', self.timerPeriod,'BusyMode','drop','ExecutionMode','fixedDelay');
            start(self.runAsyncTimer); 
            pause(1);
        end
                
        function Stop(self)
            if ~isempty(self.runAsyncTimer) && strcmp(self.runAsyncTimer.Running,'on')
                stop(self.runAsyncTimer);
            end
        end
        
        
%% Plot
% Starts a plot timer to keep updating a figure
        function Plot(self)        
            if isempty(self.runAsyncTimer) || strcmp(self.runAsyncTimer.Running,'off')
                warning('The update data timer is not currently started. Please start if you want to see new data.') %#ok<WNTAG>                
            end
            self.plotTimer = timer('TimerFcn', @(src,event)UpdatePlot(self), 'name', 'UpdatePlotTimer','Period', self.plotTimerPeriod,'BusyMode','drop','ExecutionMode','fixedDelay');
            start(self.plotTimer);
        end
        
% Stops the plot timer
        function StopPlot(self)
            if ~isempty(self.plotTimer) && strcmp(self.plotTimer.Running,'on')
                stop(self.plotTimer);
            end
        end
        
% The update plot function, which should be changed to show what needs to be shown        
        function UpdatePlot(self)
            if isempty(self.fig_h)
                self.fig_h = figure;
            end
            set(0,'CurrentFigure',self.fig_h)
            surf(self.data);
        end
            
                    
%% Record
% This will block until there is recordSeconds worth of new data then it
% will save this to file (as .mat)
        function recordFilename = Record(self,recordSeconds_local)
            if ~isempty(self.runAsyncTimer) && strcmp(self.runAsyncTimer.Running,'on')
                warning('You should not run the timer and record simultaneously. Stopping the timer') %#ok<WNTAG>
                self.Stop()
            end
            
            if nargin < 2
                recordSeconds_local = self.recordSeconds;
            end
            
            starttime = clock;
            originalDesiredSampleCount = recordSeconds_local * self.sampFreq;
            remainingDesiredSampleCount = originalDesiredSampleCount;
            
            recordData = zeros(recordSeconds_local * self.sampFreq, self.channelCount);
            currentIndex = 1;
            while remainingDesiredSampleCount > 0 
                % Make sure it breaks out in 2* the correct time even if the number of samples is not there
                if etime(clock,starttime) > 2 * recordSeconds_local
                    break;
                end
                nSamplesTaken = UpdateData(self);
                if nSamplesTaken == 0 
                    continue;
                end
                % Note that this matrix WILL PROBABLY be bigger than the desiredSampleCount 
                overShoot = currentIndex + nSamplesTaken - originalDesiredSampleCount;
                if overShoot <= 0
                    recordData(currentIndex:currentIndex+nSamplesTaken-1,:) = self.data(1:nSamplesTaken,:);                    
                    remainingDesiredSampleCount = remainingDesiredSampleCount - nSamplesTaken;
                else
                    recordData(currentIndex:originalDesiredSampleCount,:) = self.data(overShoot:nSamplesTaken,:);                    
                    remainingDesiredSampleCount = 0;
                end
                
                currentIndex = currentIndex + nSamplesTaken;                
                
                %disp(['Recording... recieved  ,',num2str(currentIndex + 1), ' samples so far in' ,num2str(etime(clock,starttime)),' secs. Still waiting for ', num2str(remainingDesiredSampleCount),' of ', num2str(originalDesiredSampleCount),' samples']);
                disp(['Recording... ',num2str(etime(clock,starttime)),' s']);
                pause(self.timerPeriod);
            end
            recordFilename = ['EEGlog',datestr(now,30),'.mat'];
            save(recordFilename,'recordData');
            disp(['Recording complete. Saved to ',recordFilename]);
        end
        
        
%% LoadRecordedData
        function LoadRecordedData(self,filename)
            if ~isempty(self.runAsyncTimer) && strcmp(self.runAsyncTimer.Running,'on')
                warning('You should not run the timer and load record data simultaneously (else it will overwrite it). I''m stopping the timmer') %#ok<WNTAG>
                self.Stop()
            end
            temp = load(filename);
            self.data = temp.recordData;
        end

        
%% UpdateData 
% Update the data matrix by pushing old values on and adding new values at
% the front
        function nSamplesTaken = UpdateData(self)
            state = calllib('edk','EE_EngineGetNextEvent',self.eEvent); 
            if state~=0 % state = 0 if everything's OK
                warning('Everything is NOT OK: Is the headset connected, turned on, and on someone''s head?'); %#ok<WNTAG>
            end
                            
            eventType = calllib('edk','EE_EmoEngineEventGetType',self.eEvent);
            %disp(eventType);
            userID=libpointer('uint32Ptr',0);
            calllib('edk','EE_EmoEngineEventGetUserId',self.eEvent, userID);

            if strcmp(eventType,'EE_UserAdded') == true
                userID_value = get(userID,'value');
                calllib('edk','EE_DataAcquisitionEnable',userID_value,true);
            end
% 
            calllib('edk','EE_DataUpdateHandle', 0, self.hData);
            nSamples = libpointer('uint32Ptr',0);
            calllib('edk','EE_DataGetNumberOfSample', self.hData, nSamples);
            
            nSamplesTaken = get(nSamples,'value') ;
            if (nSamplesTaken ~= 0)
                % Should check this and only get data if full amount is  available and new 
                % (it is required for the get call to know how big dataPtr is)
                if self.showDebug  
                    disp(['nSamplesTaken = ',num2str(nSamplesTaken)])
                end
            
                % Assign memory for the number (in lib) for samples taken on a single channel
                channelDataPtr = libpointer('doublePtr',zeros(1,nSamplesTaken));
                
                % Push the old values of the end, move newest ones to end
                self.tempData = [zeros(nSamplesTaken,self.channelCount) ...
                                ;self.data(1:self.dataRowCount-nSamplesTaken,:)];
                            
                % Go through each of the channels and get all the samples for that channel
                for i = 1:length(fieldnames(self.EE_DataChannels_enum))
                    calllib('edk','EE_DataGet',self.hData, self.DataChannels.([self.DataChannelsNames{i}]), channelDataPtr, uint32(nSamplesTaken));
                    channelData = get(channelDataPtr,'value');
                    % Fill in our matrix of the latest data                    
                    self.tempData(1:nSamplesTaken,i) = channelData;
                    
                    if (nSamplesTaken~=length(channelData))
                        warning('The nSamplesTaken shoul == length(channelData), Ask Gavin'); %#ok<WNTAG>
                    end
                end
                
                % Required so data variable is safely available externally to work with,
                % Only update the data once the samples from the most recent read operation have been added
                self.data = self.tempData;
            end 
        end
        
        
%% runSMILE
        % Run loop to collect and analyze data
        function [] = runSMILE(self, mode)
            % Check input
            mode = lower(mode);
            validatestring(mode, {'learn', 'test', 'run','demo'}, mfilename, 'mode', 1)
            
            % Ask to import old data
            icanhaz = lower(input('Would you like to import previous learning data? (y/n): ','s'));
            while ~strcmp(icanhaz, {'y', 'n'})
                disp('Please answer y or n.');
                icanhaz = lower(input('Would you like to import previous learning data? (y/n): ','s'));
            end
            
            % Import old data and write to data structures in class, overwriting current data
            if icanhaz == 'y'
                % Import data
                [filename, pathname] = uigetfile('*.xlsx', 'Pick a previous data file');
                [oldData, oldResponses] = xlsread(fullfile(pathname, filename));
                self.bandSize = 22 / size(oldData, 2);
                self.ctree = ClassificationTree.fit(oldData, oldResponses);
                
                % Reset data structures
                self.responses.clear();
                self.classifyData.clear();
                for i = 8 : self.bandSize : 30 - self.bandSize
                    fRange = sprintf('%.1f - %.1f Hz', i, i + self.bandSize);
                    self.classifyData.put(fRange, java.util.ArrayList);
                end
                
                % Add data
                for i = 1:numel(oldResponses)
                    index = 0;
                    self.responses.add(oldResponses{i});
                    for j = 8 : self.bandSize : 30 - self.bandSize
                        index = index + 1;
                        fRange = sprintf('%.1f - %.1f Hz', j, j + self.bandSize);
                        self.classifyData.get(fRange).add(oldData(i, index));
                    end
                end
            else
                % Prepare HashMap of data
                self.responses.clear();
                self.classifyData.clear();
                for i = 8 : self.bandSize : 30 - self.bandSize
                    fRange = sprintf('%.1f - %.1f Hz', i, i + self.bandSize);
                    self.classifyData.put(fRange, java.util.ArrayList);
                end
            end
            
%-----------% Execution loop
            running = true;
            figure(1), hold on
            while running
                lastFilename = self.Record(10);
                running = self.analyzeData(lastFilename, mode);
            end
            hold off
            
            if strcmp(mode, {'learn', 'test'})
                % Write learned data to an excel file
                measures = self.responses.size();
                respCell = cell(measures, 1);
                dataCell = cell(measures, length(8:self.bandSize:29));
                for i = 1:measures
                    index = 0;
                    respCell{i} = self.responses.get(i - 1);
                    for j = 8:self.bandSize:29
                        index = index + 1;
                        fRange = sprintf('%.1f - %.1f Hz', j, j + self.bandSize);
                        dataCell{i, index} = self.classifyData.get(fRange).get(i - 1);
                    end
                end
                fullCell = [respCell, dataCell];
                [filename, pathname] = uiputfile('*.xlsx', 'Save decision tree data to an excel file');
                filepath = fullfile(pathname, filename);
                if exist(filepath,'file')
                    delete(filepath);
                end
                xlswrite(filepath, fullCell);

                % Create new classification tree
                numData = cell2mat(dataCell);
                predictors = cell(22 / self.bandSize, 1);
                for i = 8 : self.bandSize : 30 - self.bandSize
                    predictors{i - 7} = sprintf('%d Hz', i);
                end
                self.ctree = ClassificationTree.fit(numData, respCell, 'PredictorNames', predictors);
                self.viewTree();
            end
        end
        
        
%% analyzeData
        % Analyze data with FFT and look for certain frequencies
        function [running] = analyzeData(self, lastFilename, mode)
            % load data
            prevData = load(lastFilename);
            
            % Channel indexes
            chan = self.EE_DataChannels_enum;
            % F7 = 5 | F3 = 6 | O1 = 10 | F4 = 14 | AF4 = 17
            left   = chan.O1;
            right  = chan.F7;
            center = chan.F3;
            
            % Time axis
            samples = size(prevData.recordData, 1);
            t = 10/samples : 10/samples : 10;
            
            % Extract channel vectors
            chanL = prevData.recordData(:, left);
            chanR = prevData.recordData(:, right);
            chanC = prevData.recordData(:, center);
            
            % Prepare frequency range of alpha and beta waves
            len   = size(chanL, 1);         % Length of signal
            next2 = 2^nextpow2(len);        % Next power of 2 from length of y
            f  = self.sampFreq / 2 * linspace(0, 1, next2 / 2 + 1)';
            a  = find(f == 8);
            ab = find(f == 12);
            b  = find(f == 30);
            
            % Calculate FFT for each channel we are interested in
            fftL = fft(chanL, next2) / len;
            fftR = fft(chanR, next2) / len;
            fftC = fft(chanC,next2) / len;
            magL = 10*log10(abs(fftL(1 : next2 / 2 + 1)));
            magR = 10*log10(abs(fftR(1 : next2 / 2 + 1)));
            magC = 10*log10(abs(fftC(1 : next2 / 2 + 1)));
            
            % Find relative quantitites, each L and R channel is adjusted by C first
            alphaRQ = (magL(a:ab) - magC(a:ab)) ./ (magR(a:ab) - magC(a:ab));
            betaRQ  = (magL(ab:b) - magC(ab:b)) ./ (magR(ab:b) - magC(ab:b));
            RQ = [alphaRQ; betaRQ];
            
            % Plot raw data on left and magnitude spectrum (happy / sad) on right
            subplot(1, 2, 1), plot(t, chanL, t, chanR), legend('Left', 'Right')
            subplot(1, 2, 2), plot(f(a:ab), alphaRQ, f(ab:b), betaRQ), legend('Alpha', 'Beta')
            
            % Valence and arousal
            alphaSumL = sum(magL(a:ab));
            alphaSumR = sum(magR(a:ab));
            betaSumL  = sum(magL(ab:b));
            betaSumR  = sum(magR(ab:b));
            arNew  = betaSumL / alphaSumL;
            valNew = betaSumR / alphaSumR;
            if (valNew >= 4.5)
                if (arNew >= 4.5)
                    emotion = 'excited/happy';
                else
                    emotion = 'angry/afraid';
                end
            else
                if (arNew >= 4.5)
                    emotion = 'calm/content';
                else
                    emotion = 'sad/depressed';
                end
            end
            fprintf('Valence and arousal analysis: %s\n', emotion)
            
%             % Plot emotional state
%             img = imread('emotion.png');
%             subplot(1,3,3), hold on
%             imagesc([0 10],[0 10], flipdim(img,1));
%             plot(arNew, valNew, 'bo', 'MarkerSize', 12);
%             set(gca, 'ydir', 'normal');
%             axis([0 10 0 10]);
%             hline = refline([0 5]);
%             set(hline,'Color','r')
            
            % Machine learning with a classification-based decision tree. If learning is on, then
            % the user is probed for their emotional state. When learning is off, the program tries
            % to predict emotional state. User input is added to data fields. When the runSMILE 
            % method ends, the new data is used to update the dicision tree. In addition, the data
            % is stored in an excel file, which can later be recalled.
            
            if strcmp(mode, 'learn') % Learning mode: collect data
                self.audio.question.play()
                feels = lower(input('Was your emotion more positive, negative, or indifferent? (p/n/i/q=quit): ','s'));
                while ~strcmp(feels, {'p', 'n', 'i', 'q'})
                    disp('Please answer p, n, i, or q.');
                    feels = lower(input('Was your emotion more positive, negative, or indifferent? (p/n/i/q=quit): ','s'));
                end
            else % Make a prediction
                bandAvg = zeros(1, 22 / self.bandSize - 1);
                index = 0;
                for i = 8 : self.bandSize : 30 - self.bandSize
                    low = find(f == i) - a + 1;
                    high = find(f == i + 1) - a;
                    index = index + 1;
                    bandAvg(index) = mean(RQ(low : high));
                end
                prediction = predict(self.ctree, bandAvg);
            end
            if strcmp(mode, {'test', 'run'}) % Don't display prediction during demo
                self.audio.(prediction{:}).play()
                fprintf('Prediction: you are feeling more %s right now\n', prediction{:});
            end
            if strcmp(mode, 'demo')
                self.demoResults.add(prediction);
            end
            if strcmp(mode, 'test') % Get user feedback in testing mode
                % Note: will not affect predictions immediately, must rerun
                % runSMILE in order for feedback to take effect currently
                self.audio.verify.play()
                vfeels = lower(input('Is this prediction correct? (y/n/q=quit): ','s'));
                while ~strcmp(vfeels, {'y', 'n', 'q'})
                    disp('Please answer y, n, or q.');
                    vfeels = lower(input('Is this prediction correct? (y/n/q=quit): ','s'));
                end

                % Match feedback
                if vfeels == 'y'
                    if strcmp(prediction, 'positive')
                        feels = 'p';
                    elseif strcmp(prediction, 'negative')
                        feels = 'n';
                    else
                        feels = 'i';
                    end
                elseif vfeels == 'n'
                    self.audio.question.play()
                    feels = lower(input('Was your emotion more positive, negative, or indifferent? (p/n/i/q=quit): ','s'));
                    while ~strcmp(feels, {'p', 'n', 'i', 'q'})
                        disp('Please answer p, n, i, or q.');
                        feels = lower(input('Was your emotion more positive, negative, or indifferent? (p/n/i/q=quit): ','s'));
                    end
                else
                    feels = 'q';
                end
            end
            if strcmp(mode, {'learn', 'test'}) % Classify data
                % Currently averages each 1 Hz frequency band in alpha and beta
                % ranges. Can make finer resolution if we wish.
                running = true;
                if feels == 'q'
                    running = false;
                elseif feels == 'p'
                    self.responses.add('positive');
                elseif feels == 'n'
                    self.responses.add('negative');
                else
                    self.responses.add('indifferent');
                end
                for i = 8 : self.bandSize : 30 - self.bandSize
                    low = find(f == i) - a + 1;
                    high = find(f == i + 1) - a;
                    bandAvg = mean(RQ(low : high));
                    fRange = sprintf('%.1f - %.1f Hz', i, i + self.bandSize);
                    self.classifyData.get(fRange).add(bandAvg);
                end
            end
        end
        
        
%% viewTree
        % Simple visualization of the current classification tree
        function [] = viewTree(self)
            view(self.ctree, 'mode', 'graph')
        end
    end
end