%This is a demo for kernel estimators

data_n = 20;
point_n = 201;
data = randn(data_n, 1);
%save sample data
load sample.mat

kernel_sigma = 0.1;
tmp = max(abs(data))+5*kernel_sigma;
x = linspace(tmp, -tmp, point_n);
mu = 0;
sigma = 1;
gaussian = exp(-0.5*((x-mu)/sigma).^2)/(sigma*sqrt(2*pi));
A = zeros(data_n, point_n);
for i = 1:data_n,
	A(i,:)=exp(-0.5*((x-data(i))/kernel_sigma).^2)/(kernel_sigma*sqrt(2*pi));
end
subplot(2,1,1);
plot(x', A'/data_n, x, sum(A)/data_n, x, gaussian, 'r:', ...
	data, zeros(size(data)), 'g*');

kernel_sigma = 0.3;
tmp = max(abs(data))+5*kernel_sigma;
x = linspace(tmp, -tmp, point_n);
gaussian = exp(-0.5*((x-mu)/sigma).^2)/(sigma*sqrt(2*pi));
A = zeros(data_n, point_n);
for i = 1:data_n,
	A(i,:)=exp(-0.5*((x-data(i))/kernel_sigma).^2)/(kernel_sigma*sqrt(2*pi));
end
subplot(2,1,2);
plot(x', A'/data_n, x, sum(A)/data_n, x, gaussian, 'r:', ...
	data, zeros(size(data)), 'g*');

%h = findobj(gcf, 'type', 'line');
%set(h, 'linewidth', 2);