kperceptron.m

%% Kernel Perceptron
%% Initialization
clear all; close all;
load data; load per index
count=0; per=0; temp=0;
[train, test, labeltr, labelte]=datapartition(fea, label,250000);
summation=0;
N1_max=2000; % Number of data points from training set
alpha=zeros(N1_max,1);
iter=1; y_hat=0;y_pred=0;
%%
for N1=2000:1000:N1_max
    for sig=0.2:0.2:0.2
        alpha=zeros(N1,1);
        N2=abs(0.04*N1); %Number of data points from testing set
        %% Learning
        for n=1:1 %Number of passes over the data
            tic
            for j=1:N1
                for i=1:N1
                    y_hat=y_hat+alpha(i,1)*labeltr(i,1)*gaussian_kernel(train(i,:),train(j,:),sig);
                end
                if(sign(y_hat)~=labeltr(j))
                    
                    alpha(j)=alpha(j)+1;
                end
                y_hat=0;
            end
            t=toc;
            
            %% Prediction
            for j=1:N2
                for i=1:N1
                    y_pred=y_pred+alpha(i,1)*labeltr(i)*gaussian_kernel(train(i,:),test(j,:),sig);
                end
                if(sign(y_pred)>=0)
                    outlabel(j,1)=1;
                else
                    outlabel(j,1)=-1;
                end
                y_pred=0;
            end
            
            %% Accuracy
            A=(outlabel==labelte(1:N2));
            count=sum(A(:));
            per(iter,1)=N1;
            if n==1
                per(iter,2)=t;
            else
                per(iter,2)=sum(per(:,2))+t;
            end
            per(iter,3)=sig;
            per(iter,4)=n;
            per(iter,5)=count/N2*100
            iter=iter+1;
            count=0;
        end
    end
end