APECSgui Installation Guide and User Manual Version 1.0 March
Contents
1. Figure 20 A sub menu for defining train sets APECSgui User Manual Version 1 0 1 Page 15 gui_genSets Number of classes 2 06A_C R1 mat Labels used 1 1 Labels present 1 1 Reshuffl les _ Reshuffle samples merce E Train set Percentage of data Percentage of data _ Validation set _ Test set Done Figure 21 Percentage of data sub menu Typing 100 for Class 1 in the sub menu for setting percentage of data means that all data from the loaded file will be used for training and will be assigned to Class 1 If you want to use only a specific time range of data for Class 1 this can be done by selecting the Time segment option For example you may wish to select data recorded only between minutes 12 and 14 of the experiment for Class 1 Figure 22 pui_genSets Number of classes 2 06A_C R1 mat Labels used 1 1 Labels present 1 1 C Reshuffle samples Rernnsnenneninrnnnrn Training set Time segments in min begin end Class1 12 14 Class2 0 0 Figure 22 Time segment sub menu For any loaded file with or without class labels the time segment procedure allows you to arbitrarily define data corresponding to class 1 and class 2 at the same time When class labels are present you can define the Validation and Test sets in a similar way Or when class labels are not present and only one Test class exists you can manual2. Define epoch s length C samples C msec C sec C min APECSgui User Manual Version 1 0 1 Page 6 Figure 5 Preprocess Module sub window for defining resampling and filter options Checking the box labeled Filter Data will allow you to choose a file containing a filter structure that works with the MATLAB fi ter m function Although you may store the filter file anywhere we advise you to keep filter files in the Filters subfolder The filter file must contain standard design dM and filter fM MATLAB filter structures APECSgui will apply the filter with a zero phase shift by forward and backward filtering the data This will effectively double the filter order We provide a MATLAB m file example_filter_design m which contains examples of how to create a few different filters with the APECSgui installation in the Filters folder Figure 6 shows example of loading a Bandstop filter saved in the Filters filter3 mat file pui_preProcessRaw ARBOITS mat Re sampling no Sampling frequency 126 Hz Filtering Bandstop Fpass 1600 2800 Direct Form FIRN Fiterstiltter3 mat Epoch length 2 see 9000 Overlap length 0 sec C Resample Data Fiter Data Figure 6 An example of loading a filter file File Filters filter3 mat represents a Bandstop filter This filter was created using the following MATLAB steps 455 filter a dM f
3. valid set classi 0x0 class 0x0 test set allClasses 2706x 11x257T Re define parameters for KPLS DOR Selected method KPLS Variables to use all Number of KPLS factors 2 Frequencies to use 0 64 Remove error samples timeEpochVariance 0 100 Center data no Kernel function Polynomial 2 0 Bins process none Plot m file no Number of KPLS factors Frequencies to use Remove error samples timeEpochVariance pera parame APECSgui User Manual Version 1 0 1 k x lt x y gt param parant o Figure 28 Selection of the second order polynomial kernel for the KPLS method Page 22 pui_runApp data_GeneratesSets seq 4 vs_16_18_ rightLANT_L CPT_subi10 mat train set class 90x 11x 256 class 90x 11x 256 valid set class1 0x0 class 0x0 test set allClasses 2719x 11x 256 Selected method PARAFAC Variables to use all Number of PARAFAC factors 2 Frequencies to use 0 64 Remove error samples timeEpochVariance 0 100 Center data no PARAFAC Opt 00000 0 PARAFAC Opt 0 0 0 PARAFAC perc remove 0 Re define parameters for PARAFAC Number of PARAFAC factors Frequencies to use Remove error samples timeEpochVariance Center data PARAFAC Options vector opt 3 plot PARAFAC Constraints vector 2 0 64 0 100 na 000000 000 Percentile removed after 1st run 0 no 2nd run i Select freq bin
4. Transforming innovations into applications APECSgui Installation Guide and User Manual Version 1 0 March 31 2012 1 Introduction APECSgui is a graphical user interface for using advanced signal processing machine learning and Statistical methods to create EEG based models of cognitive states The models are based on extensive research by PDT scientists and their partners on developing algorithms for Advanced Physiological Estimation of Cognitive Status or APECS APECSgui allows non experts who are generally familiar with EEG signal processing and human cognition to design train validate and test APECS models It also allows users to export model parameters for real time estimation of cognitive status using special purpose applications for BCI2000 www bci2000 org such as the PDT Gaze Contingency Task 2 System Requirements MATLAB Version 7 10 R2010b or higher is a pre requisite for using APECSgui and must be installed or accessible from a server on the computer you will use to run APECSgui For standard functions APECSgui does not require any MATLAB toolboxes Generally the system requirements for MATLAB see www mathworks com will determine the minimum requirements for using APECSgui These include 1 any Intel or AMD x86 processor supporting SSE2 instruction set 2 1 GB disk space for MATLAB only 3 4 GB disk space for a typical installation 1024 MB RAM minimum at least 2048 MB RAM recommended However for
5. Labels used 1 1 Labels present 1 1 _ Reshuffle samples C Train set _ Validation set Test set Figure 19 The window for setting Train Validation and Test sets Suppose you have separate data files of EEG for high and low workload periods Each file uses only one class label because it has only one workload level e g class labels either 1 or 1 If the first file is low workload when you load it you will define the low workload class only When you add a high workload file in a later step you will define the high workload class Note that in the simple case of two different classes separated into different files you can avoid using labels at all In this case after loading a file with data of one condition you can assign all the data in that file to one class only For example by checking the box labeled Train set in Figure 19 a sub menu for defining the training set will open Figure 20 There are two options for defining classes either by percentage of points Percentage of data or by time Time segment Say you want to use all data of the loaded file as training data of class 1 You can do this by clicking Percentage of data to open the sub menu for setting percentage of data Figure 21 eui_genSets fc a Number of classes 2 06A_C_R1 mat Labels used 1 1 Labels present 1 1 O Reshuffle samples Train set _ Percentage of data C Time segment _ Validation set Test set Done
6. Overlap length 0 sec Used variables all Re referencing no Select variables to use FC3 FP2 F7 FP1 HEOG VEOG on off A F gt lt Select al Deselect al APECSs Page 8 Figure 8 PreProcess Module sub window for variable selection Checking the box labeled Data re referencing will provide you options for re referencing the EEG electrode montage Figure 9 Currently available options include average reference and the average reference with global field power normalization APECSgui re references the montage using the electrodes you selected in the variable selection step only pui_preProcessRaw ARBO01TS mat Re sampling no Sampling frequency 128 Hz Filtering Bandstop Fpass 1600 2800 Direct Form Be a a ee mat nits FIR Filters filter3 mat Epoch length 2 sec wwwwnnnnnnnnnnnnnn Overlap length 0 sec Used variables all Re referencing Average Reference _ Process Events C Select subset of variables Data re referencing Average Reference Average Reference amp GF lt Figure 9 PreProcess Module sub window for data re referencing After you click the next window Figure 10 allows you to correct defined values by clicking the button Correct Values or to finish definition of the parameters by clicking the button labeled Done By clicking the ARBO01TS mat Re sampling no Sampling frequency 128 Hz Filtering Bandstop Fpass 16
7. IF WE HAD BEEN ADVISED OF THE POSSIBILITY OF SUCH DAMAGES COPYRIGHT 2012 PACIFIC DEVELOPMENT AND TECHNOLOGY LLC PALO ALTO CA USA APECSgui User Manual Version 1 0 1 Page 30 References and Endnotes APECSgui was supported by contracts Advanced Physiological Estimation of Cognitive Status APECS and Neurosensory Optimization of Information Transfer NOIT from the US Army Research Office and the Army Research Laboratory monitored by Elmar T Schmeisser and Anthony Ries Partners in the research leading to APECSgui have included Paul Nunez of Cognitive Dissonance LLC and Eran Zaidel and Andrew Hill of UCLA Additional data for modeling cognitive status were provided by NASA Ames Research Center and the US Air Force Human Engineering Lab at Wright Patterson AFB courtesy of Glenn Wilson and Chris Russell MATLAB is a product of The Mathworks Inc Natick MA www mathworks com To design special digital filters for pre processing EEG recordings and compute advanced power spectral density or optional coherence estimates APECSgui uses functions available in the optional MATLAB Signal Processing Toolbox www mathworks com products signal Users familiar with digital filter design and Matlab structures can construct such filters without this toolbox Bro R 2001 The N way Toolbox Tools for fitting multi way tensor models such as PARAFAC Updated 20 Mar 2012 http www mathworks com matlabcentral fileexchange 1
8. Meany trainFiletInd valFiletInd testFiletInd train Ytest train test factors diagonality varian lew test temp tom Figure 33 res structure for PARAFAC 1x451 double 0 0235 6s7vx1l double 637x41 double 63Fx1 double 637x11 double 6s xlilx41 double 637x11x41 double i 637x44 double 1l1lx4 double 3 2 180 i 637x41 double 637x11 double 637x44 double 41x4 double iixi double 1ixi double 41x1 double 41x1 double PCA The structure of res for PCA is similar to PARAFAC now we work with matrices instead of multi way arrays Two new variables res eigValues and res expVarPCA storing eigenvalues and information about explained variance are added The PARAFAC model associated diagnostic variables res diagonality res varian and res lev are not present in the res structure for PCA Extracting BCI2000 weights and or exporting weights to an Excel file APECSgui User Manual Version 1 0 1 Page 27 For NPLS and linear KPLS you can save results in a text file which can be loaded as a classification filter matrix within BCI2000 The output of this classifier will be a single variable indicating low versus high fatigue or workload Once you have saved results of NPLS or KPLS that you want to save as BCI2000 weights you can check the box labeled Save weights BCI2000 xls in the right bottom corner of the APECSgui panel This will ope
9. regression or classification Folded two way PCA principal components analysis KPLS kernel partial least squares PARAFAC parallel factor analysis NPLS N way partial least squares Table 2 The four modeling methods offered by the Application Module In the main APECSgui input window Figure 2 click Step 4 Application Module itto open the initial file and parameter specification window e g Add File and Load Parameters Figure 3 Again as in the previously described modules you can load either an existing parameter file or add a new file The input for the Application Module is a single file with training and testing sets created within the TrainValTest module The software automatically recognizes if the structure is a multi way array or a matrix If more than one variable e g more than one EEG electrode and more than one feature e g more than one frequency in the spectrum or coherence features are used APECSgui sets the multi array structure by default You can suppress this option by checking the box labeled no next to the Run multi way mode Option Figure 25 Suppressing the default multi way mode for array data will cause the Application Module to concatenate all existing arrays into a matrix samples x all variables and features suitable for folded analysis Fui_runApp data_GenerateSets_ 1lelec_block1_ 2 _ws_blockS 6 test ALL mat train set class1 5308e 11 257T class 4211 11 257T
10. support an EEG based model for estimating or classifying cognitive status or simply to explore the multivariate or multi way structure of your data to help generate new models The Application Module provides four different procedures Table 2 First there are two unsupervised learning modeling procedures principal components analysis PCA and the multi way PARAFAC or parallel factor analysis method Second there are two supervised learning modeling procedures supervised learning classification procedures Kernel Partial Least Squares KPLS and multi way PLS NPLS The unsupervised and supervised methods can be divided into conventional folded analyses in which some dimensions are combined or folded for analysis e g frequency and electrode or true multi way method of handling high dimensional data NPLS and PARAFAC operate on multi array tensor data while KPLS and PCA are methods operating on two dimensional arrays matrices Currently only KPLS can perform nonlinear classification all other methods are linear KPLS and PARAFAC methods served APECSgui User Manual Version 1 0 1 Page 18 as key methods in the initial development of the APECS algorithms More recently the NPLS method has proved useful for creating normative EEG based models for classification of mental workload and cognitive fatigue Unsupervised Learning exploratory analysis or Supervised Learning Mathematical Structure dimensionality reduction
11. the results The results file is a binary MATLAB file mat which contains structures named param and res The structure param holds all important parameters selected within the Application Module You can inspect these parameters can by loading the saved results file into MATLAB and by typing param in the MATLAB command Window Results stored in the res structure depend on the method used In the next section we explain this structure and differences between the methods NPLS The results res structure after running the NPLS method is depicted in Figure 32 First res meanxX and res meanyY represent mean values for the used X a Y data These variables are present only in the case where centering of data was selected Figure 26 In this particular example we worked with 11 electrodes and 49 spectral lines creating an 11 49 539 dimensional X data mean vector res meanx Y data were represented by one dimensional vector of class labels and a single mean value is stored in res meany The next three variables res YtrainFilelInd res YvalFileInd and res YtestFilelnd represent numeric vectors of the length of training validation and test data These numeric vectors index a file from which a corresponding sample epoch was taken For example res Ytrainfilelnd 1 112222233 would mean that the first three epochs of training data are from the first loaded file in the TrainValTest module the next four epochs from the second file etc Information store
12. 00 2800 Direct Form wo n FIR Filters filter3 mat Epoch length 2 sec wrwwnnnnnnnnnn 2 Overlap length 0 sec Used variables F3 FZ F4 C3 CZ FC3 FP2 F7 FP1 Re referencing Average Reference Correct Values button all selected parameters will be ignored and you will be moved to the initial parameters APECSgui User Manual Version 1 0 1 Page 9 selection window pui_preProcessRaw am x ARB01TS mat Re sampling no Sampling frequency 128 Hz Filtering no Epoch length 2 sec Overlap length 0 sec eee _ Resample Data _ Filter Data Figure 4 Figure 10 Correct Values and Done window After clicking the window depicted in Figure 11 will be open You can add a new file by clicking Add File or add file with the same set of parameters by clicking Add File with existing param By clicking Add File with existing param a sub window with a list of all currently selected files will open Here you need to select a file with parameters you want to duplicate Then the next window will open allowing you to select multiple files you can load The same set of parameters will be assigned to these files For example if you have several different files that you want to process in the same way same parameters for epoching re sampling variables selection etc you would start with a single file and repeat all steps defined above until you get to window depict
13. 088 the n way toolbox BCI2000 is a free general purpose software framework for brain computer interface BCI research and applications Documentation executables and source codes may be obtained from the BCI2000 website www bci2000 org i Trejo L J amp Rosipal R 2012 EEtrac System Installation and User Manual Version 1 0 Pacific Development and Technology LLC www pacdel com MATLAB stores matrices in column major order with subscripting of rows x columns For example a matrix D containing s 1000 samples of EEG x e 20 electrodes per channel will would have dimensions size D s e 1000 20 i APECSgui assumes that EEG samples are measured in units of microvolts While APECs models are insensitive to the units of EEG samples graph labels may be incorrect if units other than microvolts are used The pwelch m and pmtm m functions for power spectral density estimation are part of the optional MATLAB Signal Processing Toolbox and are not required for standard power spectral density estimation in APECSgui which you may compute with the fft m function Click on the links to these functions in the text for more information or see the help pages at http www mathworks com help toolbox signal Selecting window type none effectively implements a rectangular boxcar window For a discussion of window types and PSD estimators and their impact on neural time series analyses see http nipy sourceforge net nitime e
14. B Command Window Type saveEEGLAB ALLEEG filename where filename is a string of the path and filename where the data in the MATLAB format will be saved Data will be saved in the raw data format needed for APECSgui see Raw Data Format section above If filename is omitted the MATLAB file with the same name as the loaded data file will be saved into the same folder from which eeglab data were loaded APECSgui User Manual Version 1 0 1 Page 28 Pui_runApp file name resHPLS_exp mat method used MPLS total number of factors 4 used variables Fa Fi F4 C3 Cz C4 Pa Pz P4 O1 O32 Model of how many factors to save Bin default Label Hz OK Save weights Figure 35 Window for saving results into BCI2000 format Note1 Within the MATLAB Command Window also the EEG structure exists and can be used i e saveEEGLAB EEG filename In this case path to the original data can t be used and if filename is omitted MATLAB format data will be saved within the current folder Note2 You can make changes of data within eeg ab prior saving to MATLAB format For example changing sampling frequency Then eeglab creates a new EEG and adds a structure item to ALLEEG So after changing data within eeg ab you can use either saveEEGLAB ALLEEG filename where indicates a number of changed data you prefer to save or saveEEGLAB EEG filename which will save the last modification of data You can always check this changes by typing A
15. LLEEG or EEG within the MATLAB Command Window Note 3 If events are present EEG event ALLEEG event the current implementation extracts events to dat Y see Raw Data Format section above at all locations given by event atency and for each events assigns its type given by event type Other locations have event type equal to O APECSgui User Manual Version 1 0 1 Page 29 LICENSE AND WARRANTY INFORMATION LIMITED SOFTWARE LICENSE AGREEMENT THE APECSgui SYSTEM SOFTWARE EXCLUDING MATLAB PUBLIC DOMAIN PORTIONS OF BCI2000 AND THE NWAY TOOLBOX IS PROTECTED BY COPYRIGHT AND INTELLECTUAL PROPERTY RIGHTS OF PACIFIC DECELOPMENT AND TECHNOLOGY LLC YOUR RIGHTS TO USE THE SOFTWARE ARE ONLY AS SPECIFIED IN THIS AGREEMENT AND IN ARL CONTRACT W911NF 11 C 0081 WE RESERVE ALL RIGHTS NOT EXPRESSLY GRANTED TO YOU IN THIS AGREEMENT YOUR USE OF THIS SOFTWARE CONSTITUTES YOUR CONSET TO THIS AGREEMENT NOTHING IN THIS AGREEMENT CONSTITUTES A WAIVER OF OUR RIGHTS UNDER U S OR INTERNATIONAL COPYRIGHT LAWS OR ANY OTHER INTERNATIONAL FEDERAL OR STATE LAW THIS AGREEMENT AUTHORIZES YOU THE STAFF AND CONTRACTORS OF THE US ARMY RESEARCH LABORATORY ARL HED TO USE THE SOFTWARE ONLY ON COMPUTERS OWNED LEASED OR OTHERWISE CONTROLLED BY YOU USE OF THE SOFTWARE ON A COMPUTER OWNED BY A THIRD PARTY WHOIS AT THAT TIME PROVIDING IT SERVICES TO YOU IS ALLOWED PROVIDED THAT YOU MAKE EVERY REASONABLE EFFORT TO ADVISE US OF THE IDENTITY OF THE THIRD PARTY AND PR
16. OVIDED THAT YOU AGREE TO BE RESPONSIBLE FOR THAT THIRD PARTY S COMPLIANCE WITH THIS AGREEMENT USE OF THE SOFTWARE ON A COMPUTER OWNED BY A THIRD PARTY WHO IS NOT AT THAT TIME PROVIDING IT SERVICES TO YOU IS PROHIBITED INSTALLATION OF THIS SOFTWARE ON A SERVER THAT ALLOWS ACCESS TO THIS SOFTWARE VIA A PUBLIC NETWORK OR THE INTERNET WITHOUT THE USE OF A PASSWORD PROTECTED SECURE PORTAL IS PROHIBITED YOU MAY MAKE COPIES OF THE SOFTWARE FOR FOR ARCHIVAL AND BACK UP PURPOSES ONLY EACH COPY OF THE SOFTWARE YOU MAKE SHALL CONTAIN THE COMPLETE EETRAC DOCUMENTATION AND THE TEXT OF THIS AGREEMENT NO WARRANTY WE PROVIDE THE APECSgui SYSTEM AND SOFTWARE IN THE HOPE THAT IT WILL BE USEFUL BUT WITHOUT ANY WARRANTY IT IS PROVIDED AS IS WITHOUT WARRANTY OF ANY KIND EITHER EXPRESSED OR IMPLIED INCLUDING BUT NOT LIMITED TO THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM IS WITH YOU SHOULD THE SYSTGEM PROVE DEFECTIVE YOU ASSUME THE COST OF ALL NECESSARY SERVICING REPAIR OR CORRECTION NO LIABILITY TO THE MAXIMUM EXTENT PERMITTED BY APPLICABLE LAW IN NO EVENT SHALL WE BECOME LIABLE TO YOU OR TO ANY OTHER PARTY FOR ANY LOSS OR DAMAGES WHETHER INDIRECT CONSEQUENTIAL PUNITIVE SPECIAL INCIDENTAL OR OTHERWISE ARISING FROM YOUR USE OF OR INABILITY TO USE THIS SYSTEM INCLUDING BUT NOT LIMITED TO DAMAGES FOR LOSS OF TIME MONEY DATA OR GOODWILL EVEN
17. ate m pui_penFeatures ARBO4TS mat Select coherence method window length window overlap Figure 14 Features Module window for selecting coherence methods and setting parameters After clicking the second parameters window Figure 15 will be open Here you can define a range of frequencies for which power or coherence values will be computed The format in which frequency sub bands are computed is also allowed e g 4 16 17 25 However we advise you to set this range as wide as possible and covering all frequencies you may later investigate within the Application Module At any time later within the Application Module you can define sub features e g subsets of frequencies from this wide range of features This will avoid the necessity of returning to the Features Module and re computing the features whenever you decide to investigate a different set of APECSgui User Manual Version 1 0 1 Page 12 frequencies By checking the box labeled Zero mean prior spectra computation each epoch will be separately centered zero mean before computing power or coherence After clicking the final window of the Features Module will open allowing you to save parameters compute features and close the module Figure 16 By clicking the Generate Features the routine for computing features is evoked At this point we recommend that you save feature file s into the Data FeaturesData folder or to another named subfolder within this
18. atenated into a 539 dimensional vector res Xfactors res Yfactors stores loadings weights of atoms factors computed by NPLS In the example above 4 atoms were computed considering 3 way form of temporal x spatial x spectral components Therefore res Xfactrors 1 is a matrix of four 637 samples long temporal atoms Similarly res Xfactrors 2 is a matrix of four spatial atoms of size 11 corresponding to 11 selected electrodes and finally res Xfactrors 3 is a matrix of four spectral atoms representing weight vectors of spectral line selected in this case 49 spectral lines res Yfactors similarly represents Y space atoms in our case Y represents a class membership variable therefore interpretation of res Yfactors is not very meaningful Similar to res Xfactors and res Yfactors the next six variables res Core res B res ypred res Ssx res Ssy and res reg represent output variables of the npls m routine from the N way toolbox You can learn more about these variables by typing help npls within the MATLAB Command Window Here we will mention res ypred which represents the training set predicted values of Y for one to number of selected factors components in this case 4 factors res ypred is a matrix where each column represents prediction values for a model consisting from one to the column number factors i e the first column are predicted values considering the first factor only the second column considering both fact
19. be defined as a vector with identical values e g all equal to 1 The header structure must contain the five variables Xsize Ysize Xlabels sampleFreq and sampleFreqUnit These variables provide data dimensions and sampling parameters For the prior example a valid header structure could be APECSgui User Manual Version 1 0 1 Page 4 gt gt header header size 166400 32 Ysize 166400 1 labels 1x32 cell sampleFreq 126 sampleFrequnit Hz where header Xsize and header Ysize reflect the sizes of dat X and dat Y respectively The cell array header Xlabels holds the variable labels or channel names e g header Xlabels F3 FE C3 If no labels are available for better data manipulation we advise generating labels such as 1 2 3 Then in the example above we would have header Xlabels 1 2 3 32 The last two variables in the header structure store the EEG sampling frequency and the sampling frequency unit label Preprocessing EEG Data Clicking Step 1 PreProcess Module in the main input window Figure 2 APECSgui will open the PreProcess main input window Figure 3 pul_preProcessRaw Add File Load Parameters Figure 3 PreProcess Module main input window Click the button labeled Add File to start a new preprocessing task starting or Load Parameters to preprocess data using previously saved parameters Clicking the
20. bership labels are present or not and opens an appropriate menu Currently the Application Module works with binary classification only First we consider the case where class labels are present For example consider the example with several workload levels e g low workload 1 high workload 1 and undefined 0 mentioned in the PreProcess Module section A window for selecting classes by label will open after you decide to use this label information Figure 17 Figure 17 Window for defining number of class es and class labels when class labels are present in the data Because the classification methods of the Application Module must be binary in the present version of APECSgui you must remove one of the three classes Say you want to discriminate low and high workload Then eliminate Class 3 undefined workload level 0 by clearing its field which sets the class 3 label equal to an empty string Figure 18 Figure 18 An example of eliminating a class by clearing its field in this case Class 3 Then after clicking a window for setting training validation and test sets will open Figure 19 At the top of the window you can see which labels are present in the currently loaded file and which labels APECSgui will use for this file Note that you can update labels and classes if you add a new file with different labels later APECSgui User Manual Version 1 0 1 Page 14 eui_genSets ama Number of classes 2 06A_C R1 mat
21. button labeled Load Parameters simply allows you to re process files using saved parameters without re entering parameters Clicking will open the standard MATLAB Open file window where you can select a stored raw data file After selecting a data file a sub window will open where you must define pre processing parameters Error Reference source not found The first two parameters pertain to epoch length and overlap These serve to tell APECSgui how to divide the continuous EEG series into a set of equal duration epochs with or without overlap You specify epoch duration and overlap in units of milliseconds seconds or minutes and APECSgui calculates the number of samples per epoch using the sampling frequency in the header Note that for continuous non epoched EEG you may set epoch length equal to the entire record with zero overlap e g only one epoch per APECSgui User Manual Version 1 0 1 Page 5 data file After defining epoch parameters you may choose to optionally resample and filter the raw EEG data pui_preProcessRaw ARBO01TS mat Define epoch s length C samples C msec C sec C min Figure 5 eui_preProcessRaw a a x ARB01TS mat Re sampling no Sampling frequency 128 Hz Filtering no Epoch length 2 sec Overlap length 0 sec _ Resample Data Filter Data Figure 4 Preprocess Module sub window for defining epoch length and overlap options pui_preProcessRaw ARBO1TS mat
22. computation of spectral and coherence features You can start the Feature Module by clicking the button within the APECSgui panel Figure 2 As in the PreProcess Module the main input window described previously Figure 3 will open You can then select either a new file Add File of epoched data or load an existing parameter file which can be further modified If you added a new file the next two windows will determine the type of computed features and its parameters For spectral features APECSgui provides three power spectral density PSD estimates Figure 13 The default method uses the built in MATLAB method fft m you may also choose optional methods pwelch m and pmtm m Figure 13 Features Module window for selecting a PSD method and setting PSD parameters You need to specify a window type none Hamming Hann and Blackman and a set of parameters which are determined by the selected spectral method this includes number of fft points or length of sub windows for Welch s method etc Currently APECSgui provides four spectrum formats These are power power log power in dB units log_ power positive log power og_power t where positivity is enforced by computing log10 p 1 where p represents power and square root of power sgrt_power APECSgui also provides three optional methods for computing coherence Figure 14 t These are based on the MATLAB signal processing toolbox procedures cpsd m mscohere m and tfestim
23. d in these variables can be then used for a plotting function to discriminate results based on the file from which samples came APECSgui User Manual Version 1 0 1 Page 25 gt Fes res Means Meany trainFiletInd walFileInd testFiletInd ROrALN Atest Ytrain Ytest factors Yfactors Core B ypred 33X Ssy req ypred test T test 1x539 double O 023547350690736 63Fx1 double 63 x1l double 637x11x49 double 63Fx11x49 double 63Fx1 double 63Fx1 double i 637x44 double iix4 double 63 7x4 double 1 1 1 i i 1 000000000000001 exex2 double 4x4 double 637 x4 double 5x2 double 5x2 double i 11x49 double 637x44 double 637x4 double 49x4 double 3x3x3 double 11x49 double 11x49 double Figure 32 res structure for NPLS 4x4x4 double 11x49 double The next variables res Xtrain res Xtest res Ytrain and res Ytest store the actual X and Y data used If centering was selected these data are zero mean processed In the depicted example we can see a multi array multi way form of X data particularly we can see that for training we used 637 epochs 11 electrodes and 49 spectral lines The same number of samples was used for testing in this case In the case of KPLS and PCA the variables res Xtrain and res Xtest would have the format of matrices of size 637 x 539 where electrodes and spectral lines are conc
24. design bandstopi Fpi Fst1 Fst2 Fp pi 4st ap2 1600 2000 2400 2600 1 80 1 8000 EM design dM equiripple save filters dM fM Click to move to the next window where you may select optional preprocessing parameters Figure 7 Checking the box labeled Process Events allows you to specify a MATLAB script for processing labeling events parameters defined in dat Y within each epoch Such scripts must be written by the user we advise you to store such scripts within the Customized M files folder If you define dat Y as a vector of events then after epoching PreProcess Module will create a new variable APECSgui User Manual Version 1 0 1 Page 7 called datY a matrix with dimensions number of epochs x samples within an epoch However if you define dat Y as a matrix you may want to store more types of events e g reaction time correctness of response misses etc then after epoching PreProcess Module will create the structure datY var ch Y where ch indexes the columns of dat Y and for each such column a matrix of the size number of epochs x samples within an epoch is created from dat Y Therefore you must process datY according to its structure datY is the only input variable for custom event based processing code Two output variables must be stored mainly a new vector with class membership say dY and classLabel a vector with unique class membership values We provide an example in which dat Y is a vector indicating various w
25. ed in Figure 11 Now instead of repeating these steps for all remaining files you can click the Add File with existing param button again to assign the same preprocessing parameters for subsequent files You may repeat this step as needed The right panel will allow you to check a list all currently loaded files and assigned parameters You can correct parameters for a specific file by clicking the Correct File button or delete a file or a set of pui_preProcessRaw Parameters Summary save to A 1 Add File ARB01TS mat sampling freq 128 Hz epoch length 2 sec overlap 0 sec filtering Bandstop Fpass 1600 2800 Direct Form FIRN Filter used variables F3 FZ F4 C3 CZ FC3 FP2 F FP1 processing events no re referencing Average Reference APECSg Page 10 File Selection Done v A lv already defined files by clicking the Delete File button Figure 11 Window for adding correcting and deleting parameters and files If all files intended for the preprocessing step were defined and parameters set click the button labeled File Selection Done to continue to the next window Figure 12 At this point we advise that you save all selected parameters by clicking Save Parameters We also advise that you save these parameters into the ParameterFiles folder or its subfolders which can be arbitrarily created by you Saving parameters allows you to load them later either for run
26. enu will open Figure 31 Here you can select several plots as well as a smoothing parameter applied to temporal plots Not all plotting types are available within all four modeling methods Only plotting options appropriate for a selected method are available For example for PARAFAC and PCA there are no prediction plots be plotControl Sea Select plots Smoothing Parameter Plot predictions 40 10 Plot atoms C Plot PSD all Plot PSD selected channels Figure 31 The plotting options window As mentioned above you can define your own plotting function which should be saved within the Customized M files and loaded as a parameter In this case the plotting options window Figure 31 will not appear but instead your defined plot routine will run The plotting function takes to input APECSgui User Manual Version 1 0 1 Page 24 parameters which are param and res MATLAB structures described next The output parameter is the variable hF which can represent a single name or a cell of figure names created within the custom based plotting function For example you may create several figures named hF fig_1 fig 2 fig_5 This can be initialized by a loop HF COF 1 15 str fig numzstr ij hFiend l str end and later within the code each figure can be called as figure Name hF f Results structure After clicking Run App Figure 30 you can name a file for saving
27. folder pui_penFeatures ARBO4ATS miat Frequency band s to store e g 0 12 14 15 ee ee ped window type none number of FFT points 756 type of spectra log power Zero mean prior spectra computation Figure 15 Frequency range definition and data centering pui_penFeatures epoch length 2 sec overlap 0 sec psd method fit ped window type none number of FFT points 256 pe Of Spectra log power req band s to compute 0 64 Go Back to File Selection zero mean no Figure 16 The final window for saving parameters computing features and closing the module Creating Sets for Model Training Validation and Testing with the TrainValTest Module The inputs for this module are feature data file s which you can combine into training validation and or testing sets The output is a single data file which is then used as an input within the Application Module described in the next section Currently the model validation step within the Application Module is not implemented however you can define a validation subset in the TrainValTest Module As for other APECSgui User Manual Version 1 0 1 Page 13 modules you start the TrainValTest Module from the APECSgui panel Figure 2 by clicking a window for adding files and loading parameters will open same as in Figure 3 After loading a new file containing the desired features the software automatically recognizes if class mem
28. lder tree containing codes and other folders for storage of data and other file types consisting of eight subfolders Customized M files Filters M files N way ParameterFiles Results and Weights BCI2000_ xls Figure 1 The APECSgui codes do not use much space so if you will use APECSgui for various unrelated analyses we recommend creating a separate root folder for each analysis For example if your study is about Cognitive Fatigue Study 1 you may create a root folder such as F Analyses Cognitive Fatigue Study 1 on hard disk F and install APECSgui there APECSgui User Manual Version 1 0 1 Page 2 Although you can change the Data folder tree arbitrarily we do not advise this because some parts of the code rely on default paths for loading and saving files The APECSgui Data Package includes sample data for tests or demonstrations To use the sample data sets extract the contents of the data package APECSguiDataPackage zip in the Data subfolder located in the APECSgui root folder you created To use APECSgui to analyze your own data you must install your data files in the Data subfolder We will describe the data subfolder and other subfolders and how to use them in the following sections 4 Starting APECSgui To start APECSgui start MATLAB and set the working directory to the root folder you created for the APECSgui installation Then run the start up script APECSgui m which should be in the root folder if the APECSgui is prope
29. ly separate two classes when you evaluate the results e g when plotting results if you know when the classes divide An example of this may be to contrast the first half APECSgui User Manual Version 1 0 1 Page 16 of a session from the second half in a mental fatigue analysis Note that you can define all three sets can Train Validation and Test in a single file using these methods When classes are defined by percentages of data APECSgui automatically checks and warns you if the overall sum exceeds 100 However the process is different when class labels are present consistency within a class is checked than when class labels are not present consistency within all data is checked When you define classes by percentages of data APECSgui uses the following procedure When class labels are not present and you might define percentages of data such as training 10 class 1 30 class 2 validation 5 class 1 25 class 2 and testing 25 for all classes Figure 23 Then starting from the first observation APECSgui will split the file in consecutive blocks pui_genSets E a x Number of classes 2 06A_C_R1 mat C Reshuffle samples Train set Percentage of data Class1 10 Class2 30 Validation set Percentage of data Class1 5 Class2 25 Test set Percentage of data All class es 25 Done with no gaps using a total of 95 of the data and leaving 5 unused Table 1 Figure 23 An example of selecting training
30. mat file which must contain two structures dat and header APECSgui User Manual Version 1 0 1 Page 3 APECSeui OX Step 1 PreProcess Module Step 2 Features Module AVY Step 3 TrainVvalTest Module Save weights BCI2000 xls Figure 2 Main input window for APECSgui showing options for 1 pre processing data 2 feature extraction 3 model training validation and testing 4 applications of models A separate Save weights option is available for generating real time model weights to use within the PDT EEtrac system such as the GazeContigencyTask The EEtrac system is built around the general purpose BCI2000 framework for brain computer interface development The Save weights option also allows you to save model weights into a Microsoft Office Excel file The dat and header structures must be created and saved in each data file you wish to analyze and stored in the Data RawData subfolder The dat structure contains two matrices dat X and dat Y where X has dimensions of samples x electrodes and Y has dimensions of samples x events data For example consider the case of recording 166400 samples of EEG from 32 electrodes and event information e g stimulus onset times response times hits misses etc during an experiment The dat structure for these data would have the form gt gt dat dat a 166400 32 double 166400x1 double aH Both X and Y can be matrices or arrays If no events are recorded Y should
31. n the initial window for loading the MATLAB results file For all linear models you can also export weight vectors to an Excel file After loading the results file an option for saving to BCI2000 and or XLS will open Figure 34 By checking the radio buttons BCI2000 and or XLS output weights formats are determined eui_SaveWeights el X file name resNPLS_exp mat method used MPLS total number of factors 4 used variables F3 FA F4 Cs CF C4 Pa PZ Pa 01 OZ Save to BCI2000 Save to Excel xls OK Figure 34 Window for saving results into BCI2000 format Then after clicking a window similar to the one depicted in Figure 35 will be open You can select the number of model factors you wish to save By default the model with maximum number of factors is selected There are two formats of BCI2000 for referencing elements of the weight vectors see BCI2000 tutorial At the moment we recommend to use Bin format select Bin radio button After clicking Save weights you will be asked to define name of the text file storing the BCI2000 classifier weights and or of an Excel file for storing weights By default these files are stored in the Weights _BCI2000_xls but the folder can be arbitrarily changed EEGLAB 2 Raw Data First open eeglab within MATLAB Then by taking the following steps File gt Load Existing Database gt select a dataset you wish to import By doing this the structure ALLEEG becomes available within the MATLA
32. ning the same preprocessing step or for modifying or deleting parameters of selected file s or adding new files Now you can click the Pre Process Data which will start the preprocessing You will be asked first into which folder preprocessed data should be stored It is advised that you save preprocessed data into the Data EpochedData folder or to any of its subfolders which can be arbitrarily created by you Go Back to File Selection allows you to return to the previous menu Figure 11 allowing you to modify parameters Finally clicking the button closes the PreProcess Module gui_preProcessRaw fe xX Parameters Summary save to 777 A 1 z Pre Process Data 064_B_R2 mat sampling freq 256 Hz epoch length 2 sec overlap 0 sec new sampling freq 64 Hz fitering Lowpass FittersWilter2 mat used variables FZ F4 F7 CZ C4 P3 PzZ P4 01 02 FP2 75 14 T6 processing events yes set VorkloadLabels m re referencing none Go Back to File Selection v LE 2 Figure 12 Pre Process module window for saving parameters running pre processing and closing the module Features computation The inputs for this module are epoched data file s created within the PreProccess Module and the pui_genFeatures ARBO4TS mat Select psd method Type of spectra Parameters for FFT of FFT points age 11 outputs are file s with computed features Currently APECSgui implements
33. of epochs in your data set For example 5 95 will exclude all epochs with variances below the 5th percentile or above the 95th percentile of the distribution of variances of all training data epochs e Center data allows you to specify centering the features in the training set zero mean Note this centering option is not available when you apply models weights in the GazeContingencyTask using BCI2000 e Select freq bins processing allows you to select the Average parameter which will compute averages across frequencies in each range you defined in Frequencies to use The default setting is none which using all frequencies without averaging You may use the Average parameter when you are interested in analyzing bands such as theta 4 8 Hz or alpha 8 12 Hz pui_runApp _ E Eg data_GenerateSets_iielec_block1_2 Selected method NPLS Variables to use all _ws_block11_127_test_ALL mat 2 nn nnn nee nn atta tata ete tata Number of NPLS factors 2 number of classes 7 Frequencies to use 0 64 Remove error samples train set classi 53058x 11x25T class timeEpochVariance 0 100 7016x 11x257T Center data no Walid set classi 0x0 class 0x0 test set allClasses 7706x 11x25T Re define parameters for HPLS he o E e S a a E NN a E a a A Number of NPLS factors 2 N Frequencies to use 0 64 Remove error samples
34. orkload levels from which we wish to create three workload classes low workload 1 high workload 1 and undefined 0 The script for this example script is in Customized M Files setWorkloadLabels m ARBO01TS mat Re sampling no Sampling frequency 128 Hz Filtering Bandstop Fpass 1600 2800 Direct Form weeennne 2 FIR Filters filter3 mat Epoch length 2 sec wewwnnnnnnnnnnna Overlap length 0 sec Used variables all Re referencing no eee C Process Events C Select subset of variables C Data re referencing Figure 7 PreProcess Module sub window for optional event processing variable subsets and re referencing Checking the box labeled Se ect subset of variables will pop up a sub window for variable selection Figure 8 At this point we advise you to select all variables you may later use in the analysis In the Application Module representing the last step of the pipeline you will have the option arbitrarily select a sub set of variables actually used for final classification or explanatory analysis In this way a necessity to return to preprocessing step every time you decide to change a subset of variables will be avoided pui_preProcessRaw E a ed ARBO1TS mat Re sampling no Sampling frequency 128 Hz Filtering Bandstop Fpass 1600 2800 Direct Form a ee eee ene eee FIR Filttersfilter3 mat Epoch length 2 sec
35. ors one and two etc In the same way we compute predicted values for testing points These are saved within the res ypred_test variable Finally res T test stores temporal components of testing set Their linear combination creates res ypred_test For better understanding of this prediction step you can type help npred within the MATLAB Command Window APECSgui User Manual Version 1 0 1 Page 26 KPLS In the case of linear KPLS the npls m is called but now without considering a multi way structure In the case of nonlinear KPLS polynomial or Gaussian kernel selected only res ypred and res ypred_test values are provided because the kernel form of PLS does not allow access to component weights PARAFAC For PARAFAC the structure of the res is similar to NPLS but in this case no prediction of Y is carried out because PARAFAC is an unsupervised data explanatory method The structure of res for PARAFAC is depicted in Figure 33 We can see that we have no Yfactors here An important variable is res Xtest_tempAtom which stores the projection of test data onto temporal atoms in other words representing temporal atoms of testing samples The model diagnostic variables res diagonality res varian and res lev are stored in the case of PARAFAC These variables represent a core consistency diagnostic type help corcond residual variance and leverage values computed while running the PARAFAC model gt FES Fes meang
36. ove error samples timeEpochVariance 0 100 Center data no Bins process none Plot m file no F Use customized plot function t Figure 27 The second windows with two parameters common to all four implemented methods In addition parameters common to all methods you must specify additional parameters for the KPLS and PARAFAC methods For KPLS these are kernel type selection and the kernel function parameters Two types of APECSgui User Manual Version 1 0 1 Page 21 nonlinear kernel are available polynomial and Gaussian pui_runApp data GenerateSets 11elec_blocki 2 _vs_block11_1 _test_ALL mat train set class 5308x 11x257T class 2016x 11x257 walid set class1 0x0 class 0x0 test set allClasses 2706x 11u25T Re define parameters for KPLS Number of KPLS factors Frequencies to use E oi Selected method KPLS Variables to use all Number of KPLS factors 2 Frequencies to use 0 64 Remove error samples timeEpochVariance 0 100 Center data no Kernel function Polynomial 2 0 Bins process none Plot m file no 2 0 64 Remove error samples timeEpochVariance 0 100 para parame o kK x y gt param 2 param1 Figure 28 depicts the case where polynomial kernel of the second order was selected pui_runApp data GenerateSets 11elec_blocki 2 _vs_block11_12_test_ALL mat train set class1 5808x 11x257 class 2016x 11x257
37. rly installed To maintain consistency with required folders for data and other files we recommend staying in the root folder over the MATLAB session in which you will use APECSgui At start up APECSgui will set the paths to all required subfolders relative to the root folder Customized M files t Data O BehavioralData EpochedData FeaturesData TS Raw Data d TrainwalTest C Filters o M files O Neway ParameterFiles O Results Weights BCIZ000_ xls APECSgui m Figure 1 Default structure of the root folder for APECSgui which contains required application code and subfolders for raw data processed data computed features results and graphs After starting APECSgui will display the main APECSgui input window There are four main modules which create a pipeline for processing data from its raw format to running an APECS model application These are 1 PreProcess Module 2 Feature Module 3 TrainValTest Module and 4 Application Module Figure 2 An additional module is available for optionally saving APECS model weights to use within PDT real time applications for BCI2000 such as the GazeContingencyTask An option for saving weights in Microsoft Office Excel file format is also available 5 Using APECSgui Raw Data Format To use APECSgui with your data you must create raw data files with a format that APECSgui can use The APECSgui raw data file is a MATLAB data file
38. s processing none Figure 29 Selection of the additional PARAFAC parameters After setting all required parameters for a particular method you can click which will bring you to the last window of the Application Module Figure 30 APECSgui User Manual Version 1 0 1 Page 23 pui_runApp data GenerateSets seg 4 ws_16_18_rightLANT_L Selected method HPLS Variables to use all ee oe Wr pe Eee ee ee soos esses SSS Se SS SESS SSS SSS Number of NPLS factors 2 number of classes 7 Frequencies to use 0 64 Plot m file no Remove error samples train set class1 90x 11x256 class 90x 11x 256 ttiimeEpochVariance 0 100 walid set class 0x0 class 0x0 Center data no test set allClasses 2719x 11x256 Bins process none Go Back Correct Values Figure 30 The final window for saving parameters running the application and closing the module Again the structure of this window is similar to the last window within the previous modules First we recommend you save the parameters to a file Save Parameters Next by hitting the Run App button you can start running the application In the first file saving menu you will be asked where to save the obtained results We recommend saving results in the Resu ts subfolder which you can find within the APECSgui root folder After you train your model and predict values on test data a plotting sub m
39. timeEpochVariance 0 100 Center data no e e e e e e e e e e Cl l l l l l l Select freq bins processing l l l l l l APECSgui User Manual Version 1 0 1 Page 20 Figure 26 Parameters common to all four implemented methods Click OK lto select parameters and another window will open where you may set two options which are also common to all modeling methods Figure 27 Checking the box labeled Select subset of variables allows you to select a subset of variables for the model Checking the box labeled Use customized plot function allows you to use a customized function to plot results The process of selecting variables is the same as for the PreProcess Module Figure 8 Because the files you created in the TrainValTest Module contain all variables you kept in the PreProcess Module you can choose any subset of these variables when you run the Application Module The second option allows you to load a specific function for plotting results We recommend saving such a function m file within the Customized M files We will describe how to code such a function and the structure of the output results below eui_runApp data GenerateSets 11elec_block1_ _vwe_block11_127_ test_ALL mat train set class 5808u 11x257T class 2016x 11x25T walid set classi 0x0 class 0x0 test set allClasses 27706xu 11x25T Selected method HPLS Humber of HPLS factors 2 Frequencies to use 0 64 Rem
40. valid set class1 0x0 class 0x0 test set allClasses 23254x4 11x257T h Select method Select task type Classification Modeling hy merging all classes into one Fun mutti vray mode F yes ce Figure 25 Switching off multi way mode APECSgui User Manual Version 1 0 1 Page 19 As mentioned above APECSgui provides for folded and multiway approaches using supervised or unsupervised modeling However several parameters are common to all four methods We can illustrate this with an example of using NPLS Figure 26 Parameters common to all four implemented methods The common parameters are e Number of factors defines the maximum number of factors in your model latent variables e Frequencies to use defines a range of frequencies to be used in units of Hz e g 4 6 for 4 6 Hz You may also define this parameter in terms of multiple frequency range For example 4 6 12 18 will select the two frequency ranges 4 6 Hz and 12 18 Hz When using multiple ranges you must insert a space between adjacent brackets of bins e Remove error samples Before computing power spectra the PreProcess Module computes the variance of the EEG time series in each epoch and stores it in the header The Application Module can optionally use this information to exclude from training any epochs with variances below or above criteria that you specify You specify these criteria as percentiles of the distribution of the variances
41. validation and test set by percentages Training Validation Set Training Set Set Validation Set Test Set Class 1 Class 2 Class 1 Class 2 all classes Unused Total 0 10 10 40 40 45 45 70 70 95 95 100 100 Table 1 Training validation and tests sets defined by percentages of data in a single file The percentages set in Figure 23 are training 10 class 1 30 class 2 validation 5 class 1 25 class 2 and testing 25 for all classes The percentages of the data in the first row colored bars follow these specifications but the overall percentages of data in the file are additive as shown in the second row For example Validation Set 1 begins at 40 of the data because Training Sets 1 and 2 were 10 30 40 of the data On the other hand when class labels are present APECSgui uses the labels to split the data for a class into sets in the order training validation and test The presence of class labels takes precedence over time and percentage splitting That means if you define a time segment then APECSgui will use only data with class labels within the segment that matching the defined class label for a set APECSgui User Manual Version 1 0 1 Page 17 Finally if you check the box labeled Reshuffle samples APECSgui will randomly permute the data prior making any data split This will destroy the time ordering of the data sequence After you define parameters for sets you can click Done which will bring
42. vents and feedback may be contingent on EEG and eye movement features estimated in real time The PDT GazeContingencyTask for example allows for experiments in which the location of stimuli on a display may be contingent on instantaneous gaze and concurrent EEG based estimates of mental fatigue or mental workload which are estimated using APECSgui model parameters EEtrac and the GazeContingencyTask are built on the Open Source BCI2000 framework for brain computer interface development To use APECSgui model parameters with EEtrac you must install the free BCI2000 framework on the computer that you will use to run EEtrac MATLAB and APECSgui do not need to be installed on the EEtrac computer A separate document explains the hardware and software requirements for EEtrac and provides instructions for running the PDT GazeContingencyTask In Section 5 Using APECSgui we explain how to export APECSgui model parameters for use with EEtrac and BCI2000 3 Installing APECSgui To install APECSgui you must choose a root folder which will contain required application code raw data processed data computed features results and graphs To work with typical EEG and cognition experiments the root folder should be on a device that has at least 30 GB of available file storage space After choosing or creating the root folder extract the contents of the APECSgui code package APECSguiCodePackage zip in the root folder This will create a default fo
43. working with most real world EEG data sets we recommend a system with at least two high speed CPUs gt 2 GHz 4 GB of RAM and 250 GB of disk space Performance of the minimal system may run into out of memory errors for large EEG data sets To avoid such errors and for noticeably faster performance we recommend a system with four 64 bit AMD or Intel processors 8 GB of RAM and a hard disk drive running at 7200 RPM or an SSD disk We also recommend a dedicated graphics processor with a resolution of at least 1280 x 1024 pixels for displaying high resolution graphs of data features and models APECSgui also requires use of the open source N way subroutines for efficient calculation of multiway methods such as PARAFAC parallel factor analysis or N PLS multiway partial least squares regression The current version Version 3 1 is included in this APECSgui installation package We have tested APECSgui on Microsoft Windows operating systems and confirmed operation on Windows XP SP3 Windows Vista Home Premium SP2 and Windows 7 Home Premium SP1 and Mac OSX In principle any operating system that supports MATLAB such as UNIX or Linux should support APECSgui but we have not tested APECSgui with these systems We designed APECSgui to provide model parameters for the PDT EEtrac system EEtrac is a unique hardware and software system for doing experiments in which EEG and eye movements are simultaneously recorded and in which task e
44. xamples multi_taper_spectral_estimation html H To use the optional coherence estimates in APECSgui also requires the MATLAB Signal Processing Toolbox i Trejo L J Rosipal R Nunez P L Advanced Physiological Estimation of Cognitive Status The 27th Army Science Conference Orlando Florida November 29 December 2 2010 gt Rosipal R Trejo L J amp Nunez P L 2009 Application of Multi way EEG Decomposition for Cognitive Workload Monitoring In Proceedings of the 6th International Conference on Partial Least Squares and Related Methods Vinzi V E Tenenhaus M Guan R eds Beijing China pp 145 149 2009 14 Trejo L J Knuth K Prado R Rosipal R Kubitz K Kochavi R Matthews B amp Zhang Y 2007 EEG based estimation of mental fatigue Convergent evidence for a three state model In Proceedings of the HCI International 2007 and Augmented Cognition International Conference Beijing China July 22 27 pp 201 211 New York Springer LNCS APECSgui User Manual Version 1 0 1 Page 31
45. you to the previously described windows for adding files and setting or correcting parameters Figure 11 and Figure 12 Note you can load the same file several times For example you could use one segment of the file to create a training set then use the whole file all data for the test set The last window is for saving parameters generating training validation and testing sets and closing the module Figure 24 Clicking on the button labeled Generate Data Sets will cause APECSgui to create a single output file containing the training validation and or test sets that you requested At this point we recommend saving this file in the Data TrainValTest folder or any of its arbitrarily named sub folders peui_gensets E x Par r r TTN O64 _C_R1 mat number of classes 2 train blocks iclass1 10 iclass2 30 validation blocks class1 5 iclass2 25 test block all classes 25 Go Back to File Selection Figure 24 The final window for saving parameters generating training validation and testing sets and closing x the module Creating and Testing Models with the Application Module Up to now you have been learning how to use APECSgui to preprocess data compute features and create sets for training validation and testing These are necessary steps to set up the structures you will need to create and test models using your data Now comes the interesting part where you can see if your data
Download Pdf Manuals
Related Search