Differences

This shows you the differences between two versions of the page.

Link to this comparison view

example:ft_realtime_classification [2017/08/17 11:21] (current)
Line 1: Line 1:
 +{{tag>​example realtime}}
 +
 +====== Example real-time classification ======
 +
 +===== Flowchart =====
 +
 +{{:​example:​realtime:​realtime_classification.png?​400}}
 +===== Example use =====
 +
 +The simplest is to try and classify the tutorial MEG dataset which is available from the ftp server. ​ More information is on the dataset is available [[/​tutorial/​shared/​dataset|here]]. That dataset contains the stimulus classes FC, IC and FIC, corresponding to trigger values 9, 5 and 3.
 +
 +<​code>​
 +cfg = [];
 +cfg.dataset ​ = '​Subject01.ds';​
 +cfg.trialfun = '​trialfun_twoclass_classification';​
 +cfg.trialdef.numtrain ​   = 20;
 +cfg.trialdef.eventtype ​  = '​backpanel trigger';​
 +cfg.trialdef.eventvalue1 = 9; % FC
 +cfg.trialdef.eventvalue2 = 3; % FIC
 +cfg.trialdef.prestim ​    = 0.3;
 +cfg.trialdef.poststim ​   = 0.7;
 +</​code>​
 +
 +The trial definition function //​trialfun_twoclass_classification // that is being used is included in the fieldtrip/​trialfun directory. Based on the code above you can already do 
 +
 +  dummy = ft_definetrial(cfg);​
 +
 +to see how the configuration and especially the trial definition looks like:
 +
 +<​code>​
 +>> dummy.trl
 +ans =
 +         ​211 ​        ​510 ​         90         ​NaN ​          0
 +        1111        1410          90           ​2 ​          0
 +        2011        2310          90         ​NaN ​          0
 +        2911        3210          90           ​2 ​          0
 +        3811        4110          90           ​1 ​          0
 +        4711        5010          90         ​NaN ​          0
 +        5611        5910          90         ​NaN ​          0
 +        6511        6810          90           ​2 ​          0
 +        7411        7710          90           ​1 ​          0
 +        ...
 +</​code>​
 +
 +The first column is the beginsample,​ the second the endsample, the third column the offset of each segment. The fourth column indicates the class of each data segment (NaN means unknown, which happens for the third trigger type in this dataset) and the fifth column whether it should be used for training (0) or testing (1).
 +
 +
 +However, here we are not interested in the trial definition for offline processing, but instead for online classification. So based on the cfg structure above, you can run
 +
 +  ft_realtime_classification(cfg);​
 +
 +The **[[reference:​ft_realtime_classification]]** function will print the classification result on screen and will open a figure in which the timing is displayed. It being an offline application here, the timing is measured relative to the amount of data that is processed. An acceleration factor larger than 1 means that data is processed faster than realtime, whereas smaller than 1 would indicate that it cannot keep up with the realtime speed. Note that there is quite some time spent on plotting the timing figure. Furthermore note that the timing is relative to the processed data, whereas there is also time between the trials for which the data does not have to be processed.
 +
 +===== Matlab code =====
 +
 +<code matlab>
 +function ft_realtime_classification(cfg)
 +
 +% FT_REALTIME_CLASSIFICATION is an example realtime application for online
 +% classification of the data. It should work both for EEG and MEG.
 +%
 +% Use as
 +%   ​ft_realtime_classification(cfg)
 +% with the following configuration options
 +%   ​cfg.channel ​   = cell-array, see FT_CHANNELSELECTION (default = '​all'​)
 +%   ​cfg.trialfun ​  = string with the trial function
 +%
 +% The source of the data is configured as
 +%   ​cfg.dataset ​      = string
 +% or alternatively to obtain more low-level control as
 +%   ​cfg.datafile ​     = string
 +%   ​cfg.headerfile ​   = string
 +%   ​cfg.eventfile ​    = string
 +%   ​cfg.dataformat ​   = string, default is determined automatic
 +%   ​cfg.headerformat ​ = string, default is determined automatic
 +%   ​cfg.eventformat ​  = string, default is determined automatic
 +%
 +% This function works with two-class data that is timelocked to a trigger.
 +% Data selection is based on events that should be present in the
 +% datastream or datafile. The user should specify a trial function that
 +% selects pieces of data to be classified, or pieces of data on which the
 +% classifier has to be trained.The trialfun should return segments in a
 +% trial definition (see FT_DEFINETRIAL). The 4th column of the trl matrix
 +% should contain the class label (number 1 or 2). The 5th colum of the trl
 +% matrix should contain a flag indicating whether it belongs to the test or
 +% to the training set (0 or 1 respectively).
 +%
 +% Example useage:
 +%   cfg = [];
 +%   ​cfg.dataset ​ = '​Subject01.ds';​
 +%   ​cfg.trialfun = '​trialfun_Subject01';​
 +%   ​ft_realtime_classification(cfg);​
 +%
 +% To stop the realtime function, you have to press Ctrl-C
 +
 +% Copyright (C) 2009, Robert Oostenveld
 +%
 +% Subversion does not use the Log keyword, use 'svn log <​filename>'​ or 'svn -v log | less' to get detailled information
 +
 +% this makes use of an external classification toolbox
 +hastoolbox('​prtools',​ 1);
 +
 +% set the default configuration options
 +if ~isfield(cfg,​ '​dataformat'​), ​    ​cfg.dataformat = [];      end % default is detected automatically
 +if ~isfield(cfg,​ '​headerformat'​), ​  ​cfg.headerformat = [];    end % default is detected automatically
 +if ~isfield(cfg,​ '​eventformat'​), ​   cfg.eventformat = [];     end % default is detected automatically
 +if ~isfield(cfg,​ '​channel'​), ​       cfg.channel = '​all'; ​     end
 +if ~isfield(cfg,​ '​bufferdata'​), ​    ​cfg.bufferdata = '​last'; ​ end % first or last
 +
 +% translate dataset into datafile+headerfile
 +cfg = ft_checkconfig(cfg,​ '​dataset2files',​ '​yes'​);​
 +cfg = ft_checkconfig(cfg,​ '​required',​ {'​datafile'​ '​headerfile'​});​
 +
 +% ensure that the persistent variables related to caching are cleared
 +clear read_header
 +% start by reading the header from the realtime buffer
 +hdr = ft_read_header(cfg.headerfile,​ '​cache',​ true);
 +
 +% define a subset of channels for reading
 +cfg.channel = channelselection(cfg.channel,​ hdr.label);
 +chanindx ​   = match_str(hdr.label,​ cfg.channel);​
 +nchan       = length(chanindx);​
 +
 +if nchan==0
 +  error('​no channels were selected'​);​
 +end
 +
 +% these are for the data handling
 +prevSample = 0;
 +count      = 0;
 +
 +% measure the timeing
 +tic;
 +t(1) = toc;
 +s(1) = 0;
 +
 +% these are for the classification
 +W           = [];
 +correct ​    = [];
 +train_class = [];
 +train_dat ​  = [];
 +clear(cfg.trialfun);​
 +
 +%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
 +% this is the general BCI loop where realtime incoming data is handled
 +%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
 +while true
 +
 +  % determine latest header and event information
 +  event     = read_event(cfg.dataset,​ '​minsample',​ prevSample+1); ​ % only consider events that are later than the data processed sofar
 +  hdr       = read_header(cfg.dataset,​ '​cache',​ true); ​            % the trialfun might want to use this, but it is not required
 +  cfg.event = event; ​                                              % store it in the configuration,​ so that it can be passed on to the trialfun
 +  cfg.hdr ​  = hdr;                                                 % store it in the configuration,​ so that it can be passed on to the trialfun
 +
 +  % evaluate the trialfun, note that the trialfun should not re-read the events and header
 +  fprintf('​evaluating ''​%s''​ based on %d events\n',​ cfg.trialfun,​ length(event));​
 +  trl = feval(cfg.trialfun,​ cfg);
 +
 +  % the code below assumes that the 4th column of the trl matrix contains
 +  % the class label and the 5th column a boolean indicating whether it is a
 +  % training set item or test set item
 +  if size(trl,​2)<​4
 +    trl(:,4) = nan; % don't asign a default class
 +  end
 +  if size(trl,​2)<​5
 +    trl(:,5) = 0; % assume that it is a test set item
 +  end
 +
 +  fprintf('​processing %d trials\n',​ size(trl,​1));​
 +
 +  for trllop=1:​size(trl,​1)
 +
 +    begsample = trl(trllop,​1);​
 +    endsample = trl(trllop,​2);​
 +    class     = trl(trllop,​4);​
 +    train     = trl(trllop,​5)==1;​
 +    test      = trl(trllop,​5)==0;​
 +
 +    % remember up to where the data was read
 +    prevSample ​ = endsample;
 +    count       = count + 1;
 +    fprintf('​-------------------------------------------------------------------------------------\n'​);​
 +    fprintf('​processing segment %d from sample %d to %d, class = %d, train = %d\n', count, begsample, endsample, class, train);
 +
 +    % read data segment from buffer
 +    dat = read_data(cfg.datafile,​ '​header',​ hdr, '​begsample',​ begsample, '​endsample',​ endsample, '​chanindx',​ chanindx, '​checkboundary',​ false);
 +
 +    %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
 +    % keep track of the timing
 +    %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
 +    t(end+1) = toc;
 +    s(end+1) = endsample;
 +
 +    % compute the cummulative and instantaneous number of samples per second
 +    % compare these to the sampling frequency to get the relative acceleration factor
 +    instantaneous = [nan diff(s) ./ diff(t)];
 +    cumulative ​   = (s-s(1)) ./ (t-t(1));
 +    semilogy([instantaneous(:​) cumulative(:​)]/​hdr.Fs,​ '​.'​);​
 +    title('​acceleration factor'​);​
 +    legend({'​instantaneous',​ '​cumulative'​});​
 +    % force Matlab to update the figure
 +    drawnow
 +
 +    %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
 +    % from here onward it is specific to the processing of the data
 +    %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
 +
 +    % apply some preprocessing options
 +    dat = preproc_baselinecorrect(dat);​
 +
 +    if test
 +      % retrain the classifier based on the accumulated training data
 +      if isempty(W) && numel(unique(train_class))==2
 +        % only if the classifier needs to be retrained and can be retrained
 +        fprintf('​retraining the classifier based on %d examples\n',​ length(train_class));​
 +        A = dataset(train_dat,​ train_class);​
 +        W = svc(A);
 +      end
 +
 +      % classify this trial
 +      if ~isempty(W)
 +        [nchan, nsmp] = size(dat);
 +        dat = reshape(dat,​ [1, nchan*nsmp]);​
 +        B   = dataset(dat,​ class);
 +        Bc  = B*W;
 +        estimate = labeld(Bc); ​         % this is the estimated class
 +      else
 +        warning('​classifier has not yet been trained'​);​
 +        estimate = nan;
 +      end
 +
 +      % keep track of the classification performance
 +      fprintf('​estimated class = %d, real class = %d\n', estimate, class);
 +      if ~isnan(class)
 +        % this can only be done if the true class is known
 +        correct(end+1) = (estimate==class);​
 +        fprintf('​classification rate = %d%%\n',​ round(mean(correct)*100));​
 +      end
 +    end % if test
 +
 +    if train
 +      % delete the previously trained classifier
 +      W = [];
 +      % add the current trial to the training data
 +      fprintf('​adding one example to the training dataset\n'​);​
 +      [nchan, nsmp] = size(dat);
 +      dat = reshape(dat,​ [1, nchan*nsmp]);​
 +      if isempty(train_dat)
 +        train_dat ​  = dat;
 +        train_class = class;
 +      else
 +        train_dat ​  = cat(1, train_dat, ​  dat);
 +        train_class = cat(1, train_class,​ class);
 +      end
 +    end % if train
 +
 +  end % looping over new trials
 +end % while true
 +</​code>​