main_figure_part4.m

% num_DL_size_points=9;
% row_list = [551, 650; 826, 925; 1101, 1200];
model_list = ["singleonline","compositiononline", "reservoir_samplingonline", "singlejoint"];
% 
% Rate_Data_Generator(1, num_DL_size_points, row_list, model_list)
% Rate_Data_Generator(2, num_DL_size_points, row_list, model_list)
% Rate_Data_Generator(3, num_DL_size_points, row_list, model_list)
% 
% 
% for k = 1:3
%     load(sprintf('pred_rate_%d.mat', k-1))
%     figure(k)
%     
%     plot(DL_size_array', DLrate(1, :), 'x-'); hold on;
%     plot(DL_size_array', DLrate(2, :), 'o-'); hold on;
%     plot(DL_size_array', DLrate(3, :), '.-'); hold on;
%     plot(DL_size_array', DLrate(4, :), 's-'); hold on;
%     
%     plot(DL_size_array', OPTrateH(1, :), '--k'); hold on;
%     plot(DL_size_array', OPTrateL(1, :), ':k'); hold on;
%     legend('TL', 'Bilevel (Proposed)', 'Reservoir', 'Joint', 'Genie-Aided', 'Baseline')
%     ylabel('Effective Achievable Rate (bps/Hz)')
%     xlabel('Number of samples seen in data stream (k)')
%     savefig(sprintf('fig_ep%d.fig', k-1))
% end

figure()
for k = 1:size(model_list, 2)
    model =model_list(k);
    loss2plot = zeros(9, 3);
    time2plot = zeros(1, 9);
    for p = 1:4
        load(sprintf('save_%s_part%d.mat', model, p-1));
        loss2plot = loss2plot + loss;
        time2plot = time2plot + time;
    end
    
    plot(time2plot', mean(loss2plot,2), 'o--'); hold on;
    
end

grid on;
legend('TL', 'Bilevel (Proposed)', 'Reservoir', 'Joint')
ylabel('Mean Squared Error (MSE)')
xlabel('Cpu Time (sec.)')
savefig('fig_time.fig')