processAnnotationData.m

% Failure identification Raw - preprocessing
% Compute the main indicators in moving window
% Autor - Petr Šopák (221022)
% 05.11.2025

function Awin = processAnnotationData(dataAnnot, valid_mask, fps)
% ---------------------------------------------------------------
% Aggregates subjective annotation features into 1 s windows
% using the same structure as processVideoData.
%
% INPUT:
%   dataAnnot  - numeric matrix with first column = time [s]
%   valid_mask - logical vector (same length as samples)
%   fps        - sampling frequency (typically 30 Hz)
%
% OUTPUT:
%   Awin.meanVals     - [K × F] matrix of mean values per feature
%   Awin.featureNames  - cellstr of feature labels
%
% ---------------------------------------------------------------

    % ---- Constants ----
    min_valid_frac = 0.8;
    W = max(1, round(fps * 1.0));  % 1 s window
    H = max(1, round(fps * 0.5));  % 50% overlap

    % ---- Extract ----
    Time_s = dataAnnot(:,1);
    Values = dataAnnot(:,2:end);

    % Prepare feature names (if loaded via readtable)
    % If readmatrix used, dummy labels
    nFeat = size(Values,2);
    featNames = "AnnotFeature_" + string(1:nFeat);

    % ---- Prepare output ----
    N = numel(Time_s);
    starts = 1:H:(N - W);
    K = numel(starts);
    meanVals = nan(K, nFeat);

    for k = 1:K
        ii = starts(k):(starts(k)+W-1);
        finiteMask = all(isfinite(Values(ii,:)),2) & valid_mask(ii);
        if mean(finiteMask) < min_valid_frac
            continue;
        end
        meanVals(k,:) = mean(Values(ii,:), 'omitnan');
    end

    % ---- Output ----
    Awin.meanVals = meanVals;
    Awin.featureNames = featNames;
end