clear;
clc;

epochs=100;
learning_rate=0.1;
hidden_nodes=80;
place=linspace(0,2*pi,hidden_nodes);
var=0.07;

%training and test sets
x=(0:0.01:2*pi);
test_set=(0.05:0.01:6.33);

%gaussian noise
variance=0.6;
noise=sqrt(variance).*randn(1,length(x));

%Target sets
t1=sin(2*x)+noise;
t2=square(2*x)+noise;

weight_matrix=rand(hidden_nodes,1);
first_weight=weight_matrix;
for i=1:epochs
    for j=1:length(x)

        %plot the output of hidden nodes
        for k=1:length(place)
        y(k,:)=rbf(test_set(k),place(k),var);
        %plot(y(k,:))
        phi=y';
        end

        out=phi*weight_matrix;
        error=t1(j)-out;
        delta=learning_rate * error * y;
        %delta
        weight_matrix=weight_matrix+delta;
    end
        permute = randperm(length(x));
        % Permute data
        x = x(:, permute);
end

    %results
    for i=1:length(place)
        a(k,:)=rbf(x,place(k),var);
        %plot(y(k,:))
        phi=a';
    end

        out1=phi*first_weight;
        out2=phi*weight_matrix;
        plot(t1,'g')
        hold on
        plot(out1','r')
        hold on
        plot(out2,'b')