run_comparison.m

clear
close all

dataset = 'covtype'
load(['datasets_clean/', dataset]);


for j=2:size(X,2)
    temp1=abs(X(:,j));
    temp=max(temp1);
        for i=1:size(X,1)
            X(i,j)=X(i,j)/temp;
        end
end

num_feature=size(X,2);
total_sample=size(y,1);

if dataset == "covtype"
    mu = 1;
    no_workers = 200;
    rho=50;
    alpha=1;
    k = 20
elseif dataset == "SUSY" 
    mu = 1;
    no_workers = 60;
    rho=50;
    alpha=1;
    k = 20
elseif dataset == "cod-rna" 
    mu = 1;
    no_workers = 60;
    rho=50;
    alpha=1;
    k = 10

elseif dataset == "phishing" % iter = 11
    mu = 1;
    no_workers = 40; 
    rho=0.1; 
    alpha=0.25;
    k = 17;
end

lambda_logistic = 1E-3;
num_iter = 100;
repeat = 3;

dataSamples_per_worker=floor(total_sample/no_workers);

total_sample =no_workers*dataSamples_per_worker;

X_fede=X;
y_fede=y;

[obj_snewton]=standard_newton...
    (X_fede,y_fede, no_workers, num_feature, dataSamples_per_worker, num_iter,lambda_logistic);
obj0 = obj_snewton(end);

for i = 1:repeat
    [obj_FedNS(:, i), loss_FedNS(:, i), ~]=FedNS...
        (X_fede,y_fede, no_workers, num_feature, dataSamples_per_worker, num_iter, obj0, lambda_logistic, k, "Gaussian", mu);

    [obj_FedNDES(:, i), loss_FedNDES(:, i), ~]=FedNDES...
        (X_fede,y_fede, no_workers, num_feature, dataSamples_per_worker, num_iter, obj0, lambda_logistic, 2*k, "Gaussian", mu);

    [obj_FedNewton(:, i), loss_FedNewton(:, i), ~]=FedNewton...
        (X_fede,y_fede, no_workers, num_feature, dataSamples_per_worker, num_iter, obj0, lambda_logistic);

    [obj_GD(:, i), loss_GD(:, i), ~]=GD...
        (X_fede,y_fede, no_workers, num_feature, dataSamples_per_worker, num_iter, obj0, lambda_logistic);

    [obj_znewton(:, i), loss_znewton(:, i), ~]=newton_zero...
        (X_fede,y_fede, no_workers, num_feature, dataSamples_per_worker, num_iter, obj0, lambda_logistic);

    [obj_newton_aadmm_Hk(:, i), loss_newton_aadmm_Hk(:, i), ~]=newton_ADMM_Hk...
        (X_fede,y_fede, no_workers, num_feature, dataSamples_per_worker, num_iter, obj0, lambda_logistic, rho, alpha);
end

h = figure(1);
semilogy(mean(loss_GD, 2),'LineWidth',2);
hold on
semilogy(mean(loss_FedNewton, 2),'LineWidth', 2);
semilogy(mean(loss_znewton, 2),'k','LineWidth', 2);
semilogy(mean(loss_newton_aadmm_Hk, 2),'LineWidth',2);
semilogy(mean(loss_FedNS, 2),'LineWidth',2);
semilogy(mean(loss_FedNDES, 2),'LineWidth',2);

xlabel({'Number of communication rounds'},'fontsize',16,'fontname','Times New Roman')
ylabel('$f(x^t) - f(x^*)$','Interpreter','latex','fontsize', 16, 'fontweight','bold')
legend({'FedAvg', 'FedNewton', 'FedNL', 'FedNew', 'FedNS', 'FedNDES'},'fontsize', 16, 'Location', 'best');
ylim([1E-13 1E4]) 
set(gca,'fontsize',14,'fontweight','bold');
print(h, ['./results/', dataset, '.pdf'], '-dpdf','-r600')

save(['./results/', dataset, '_data.mat'],'loss_znewton', 'loss_newton_aadmm_Hk', 'loss_GD', 'loss_FedNewton', 'loss_FedNS', 'loss_FedNDES')