ICL_feature_extractor_par.m

% extract features for the ICLabel Classifier
% if there are any issues, report them to lpionton@ucsd.edu

function features = ICL_feature_extractor_par(EEG, flag_autocorr)
%% check inputs
if ~exist('flag_autocorr', 'var') || isempty(flag_autocorr)
    flag_autocorr = false;
end
ncomp = size(EEG.icawinv, 2);

% check for ica
assert(isfield(EEG, 'icawinv'), 'You must have an ICA decomposition to use ICLabel')

% assuming chanlocs are correct
if ~strcmp(EEG.ref, 'averef')
    [~, EEG] = evalc('pop_reref(EEG, [], ''exclude'', setdiff(1:EEG.nbchan, EEG.icachansind));');
end

% calculate ica activations if missing and cast to double
if isempty(EEG.icaact)
    EEG.icaact = eeg_getica(EEG);
end
EEG.icaact = double(EEG.icaact);

% check ica is real
assert(isreal(EEG.icaact), 'Your ICA decomposition must be real to use ICLabel')

%% calc topo
topo = zeros(32, 32, 1, ncomp);

%haspar = [];
%haspar = ver('parallel');
% 
% if ~isempty(haspar)
%     parfor it = 1:ncomp
%         if ~exist('OCTAVE_VERSION', 'builtin')
%             [~, temp_topo, plotrad] = ...
%                 topoplotFast(EEG.icawinv(:, it), EEG.chanlocs(EEG.icachansind), ...
%                 'noplot', 'on');
%         else
%             [~, temp_topo, plotrad] = ...
%                 topoplot(EEG.icawinv(:, it), EEG.chanlocs(EEG.icachansind), ...
%                 'noplot', 'on', 'gridscale', 32);
%         end
%         temp_topo(isnan(temp_topo)) = 0;
%         topo(:, :, 1, it) = temp_topo / max(abs(temp_topo(:)));
%     end
% else
    for it = 1:ncomp
        if ~exist('OCTAVE_VERSION', 'builtin')
            [~, temp_topo, plotrad] = ...
                topoplotFast(EEG.icawinv(:, it), EEG.chanlocs(EEG.icachansind), ...
                'noplot', 'on');
        else
            [~, temp_topo, plotrad] = ...
                topoplot(EEG.icawinv(:, it), EEG.chanlocs(EEG.icachansind), ...
                'noplot', 'on', 'gridscale', 32);
        end
        temp_topo(isnan(temp_topo)) = 0;
        topo(:, :, 1, it) = temp_topo / max(abs(temp_topo(:)));
    end
%end
% cast
topo = single(topo);
    
%% calc psd
psd = eeg_rpsd(EEG, 100);

% extrapolate or prune as needed
nfreq = size(psd, 2);
if nfreq < 100
    psd = [psd, repmat(psd(:, end), 1, 100 - nfreq)];
end

% undo notch filter
for linenoise_ind = [50, 60]
    linenoise_around = [linenoise_ind - 1, linenoise_ind + 1];
    difference = bsxfun(@minus, psd(:, linenoise_around), ...
        psd(:, linenoise_ind));
    notch_ind = all(difference > 5, 2);
    if any(notch_ind)
        psd(notch_ind, linenoise_ind) = mean(psd(notch_ind, linenoise_around), 2);
    end
end

% normalize
psd = bsxfun(@rdivide, psd, max(abs(psd), [], 2));

% reshape and cast
psd = single(permute(psd, [3 2 4 1]));

%% calc autocorrelation?
if flag_autocorr
    if EEG.trials == 1
        if EEG.pnts / EEG.srate > 5
            autocorr = eeg_autocorr_welch(EEG);
        else
            autocorr = eeg_autocorr(EEG);
        end
    else
        autocorr = eeg_autocorr_fftw(EEG);
    end

    % reshape and cast
    autocorr = single(permute(autocorr, [3 2 4 1]));
end

%% format outputs
if flag_autocorr
    features = {0.99 * topo, 0.99 * psd, 0.99 * autocorr};
else
    features = {0.99 * topo, 0.99 * psd};
end