SS_WeightFeaturesInSubset_iterate.m

function errors = SS_WeightFeaturesInSubset_iterate

% load('run_options.mat');
% load('HCTSA_N.mat');
% load('linkage_clustered_ops.mat');
% load('auto_chosen_ops.mat');

% Load in data:
HCTSA_loc = '/Users/carl/PycharmProjects/DimRedHCTSA/Matlab/data/Empirical_realizations=1000__params/fromCluster/HCTSA_N.mat';
[TS_DataMat,~,Operations] = TS_LoadData(HCTSA_loc); % 'raw');

% calculate distances between TS
dists = pdist(TS_DataMat); % TS_DataMat); %

% filter by name
% wantedFeatureNames = { ...
%     'CO_Embed2_Basic_tau.incircle_1', ...
%     'CO_Embed2_Basic_tau.incircle_2', ...
%     'FC_LocalSimple_mean1.taures', ...
%     'SY_SpreadRandomLocal_ac2_100.meantaul', ...
%     'DN_HistogramMode_10', ...
%     'SY_StdNthDer_1', ...
%     'AC_9', ...
%     'SB_MotifTwo_mean.hhh', ...
%     'EN_SampEn_5_03.sampen1', ...
%     'CO_FirstMin_ac', ... 'first_min_acf' replaced
%     'DN_OutlierInclude_abs_001.mdrmd', ...
%     'CO_trev_1.num', ...
%     'FC_LocalSimple_lfittau.taures', ...
%     'SY_SpreadRandomLocal_50_100.meantaul', ...
%     'SC_FluctAnal_2_rsrangefit_50_1_logi.prop_r1', ...
%     'PH_ForcePotential_sine_1_1_1.proppos', ...
%     'SP_Summaries_pgram_hamm.maxw', ...
%     'SP_Summaries_welch_rect.maxw', ...
%     };

% those features are now sorted by mean execution time over the 1000
% empirical TS
sortedFeatureNames = {...
    'PH_Walker_prop_01.sw_propcross', ...          
    'CO_trev_1.num', ...        
    'DN_HistogramMode_10', ...
    'SY_StdNthDer_1', ...
    'SB_MotifTwo_mean.hhh', ...
    'SY_SpreadRandomLocal_50_100.meantaul', ...
    'SC_FluctAnal_2_rsrangefit_50_1_logi.prop_r1', ...
    'AC_9', ...
    'FC_LocalSimple_mean1.taures', ...
    'CO_FirstMin_ac', ...
    'CO_Embed2_Basic_tau.incircle_2', ...
    'CO_Embed2_Basic_tau.incircle_1', ...       
    'DN_OutlierInclude_abs_001.mdrmd', ...        
    'FC_LocalSimple_lfit.taures', ...
    'SY_SpreadRandomLocal_ac2_100.meantaul', ...
    'EN_SampEn_5_03.sampen1'};
 
errors = zeros(numel(sortedFeatureNames),1);
for numFeatures = 1:numel(sortedFeatureNames)
    
    wantedFeatureNames = sortedFeatureNames(1:numFeatures);

    OperationIDs = [Operations.ID];
    OperationNames = {Operations.CodeString};

    % indices of the wanted features
    wantedOpInds = find(cellfun(@(x) ismember(x, wantedFeatureNames), OperationNames));

    redOps = Operations(wantedOpInds);

    % Create a reduced data matrix using the automatically chosen operations
    reducedMat = TS_DataMat(:,wantedOpInds);

    n = numel(wantedOpInds);
    w0 = ones(1,n);

    fun = @(w) sum(sum(abs(pdist(reducedMat.*repmat(w,size(reducedMat,1),1)) - dists)));

    options = optimset('PlotFcns',@optimplotfval);
    [w, err] = fminsearch(fun,w0,options)
    
    errors(numFeatures) = err;
end

figure
plot(1:numel(sortedFeatureNames), errors, 'LineWidth', 2)
xlabel('num features')
ylabel('error (sum of absolute differences in distances between TS)');

% % calculate distances between TS
% reducedDists = pdist(reducedMat.*repmat(w,size(reducedMat,1),1)); % TS_DataMat); % 
% 
% % Cluster the time series using the reduced (K-medoids) operation space
% maxIter = 100;
% av_ts_cluster_size = 100; % cluster count can't be more than nTS/nFeatures?
% ts_km_repeats = 10;
% ts_k = round(size(TS_DataMat,1) / av_ts_cluster_size);
% 
% [~,Cass,~,Cord] = BF_kmedoids(squareform(reducedDists), ts_k, maxIter, ts_km_repeats); 
%     
% % Sort clusters by size
% [~,I] = sort(cellfun(@length,Cord),'descend');
% 
% figure;
% plotOptions	= struct('plotFreeForm',1,'displayTitles',0,'newFigure',0); 
% numPerGroup = 25;
% maxLength = 500;
% numPlots = min(length(I),12);
% for i = 1:numPlots
%     subplot(2,numPlots / 2,i);
%     series = find(Cass == I(i));
% 
%     TS_plot_timeseries(HCTSA_loc,numPerGroup,series,maxLength,plotOptions);
%     title(sprintf('Cluster %i (%i time series)',i,length(series)));
% end

% % Save image
% fName = ['ts_clusters_',num2str(ks(kIdx)),'_',num2str(corr_dist_threshold),'.png'];
% set(gcf, 'Position', get(0, 'Screensize'));
% print(fName,'-dpng')