saker

addpath '/Users/noahakesson/MATLAB-Drive/Comp2/Tidserie/functions'
addpath '/Users/noahakesson/MATLAB-Drive/Comp2/Tidserie/data'
addpath '/Users/noahakesson/MATLAB-Drive/Comp2/Tidserie/projDatafiles'
%%
clc
clear all
close all

%% debugging creating own data
% For debugging, try generating data that has similar characteristics.
% N = 10000; extraN = 100; time = 1:N;
% C  = [ 1 ];
% A = conv([1 -1],[ 1 0.37 ]);
% e  = randn( N+extraN, 1 );
% data = filter( C, A, e );     data = data(extraN+1:end) + 1000;
% figure
% plot(data)
% noLags = 100;

%%
load('utempSla_9395.dat')
date=utempSla_9395(:, 1);
hour=utempSla_9395(:, 2);
data=utempSla_9395(:, 3);

%% behövs bara för att ta fram datum
%sve_date9395 = utempSla9395.VarName1;

%% interpolate data
% Remove and interpolate the zeros
data = remove_and_interpolate_zeros(data);

% Print the interpolated data
plot(data(10000:11680))

%%

%M-data
l=24*7*10;
start_M=(length(data)/2)+140;
last_M=start_M+l;

review_Kstad(data(start_M:last_M), 200)

%review_Kstad(utempSla_9395(:, 3), 200)

%V-data
l2=24*7*2;
start_V=last_M;
last_V=start_V+l2;

%Hoppa över-data
l3=24*7*25;
start_T=last_V;
last_T=start_T+l3;

%Test-data
l4=24*7*2;
start_T=last_T;
last_T=start_T+l4;


review_Kstad(data(start_T:last_T), 200)

%% creat arrays containing the different sets of data

data_modeling = data(start_M:last_M);
data_validation = data(start_V-24:last_V);
data_prediction = data(start_T:last_T);

data_comb_mod_val = data(start_M:last_V)

%% Lets form the k-step prediction. k=1
% Define the number of steps to look back
stepsBack = 24;

% Create an array to hold the predicted values
yhatk = zeros(length(data_validation),1);

% Loop through the input data and make predictions
for i = 1:length(data_validation)
    % If we are on the first few values, we can't predict anything
    if i <= stepsBack
        yhatk(i) = 0;
    else
        % Make the prediction by using the value from 24 steps back
        yhatk(i) = data_validation(i - stepsBack);
    end
end

% Print the predictions
yhatk

% Predict future values

%data_validation_differentiated_temp = data_validation_differentiated(24:end)

data_validation = data_validation(25:end);
yhatk = yhatk(25:end)

ehat = data_validation - yhatk;
checkIfWhite( ehat );
var(ehat)

%% Lets have a look at the prediction, comparing it to the true data.
% Looking at the figure, note that:
%   1) The prediction appears to be shifted k step to the data (it should be, so this is correct).
%   2) The inital k predicted values are zeros due to the zeros in the G polynomial.
k=1
noLags = 100
figure
plot(data_validation)
hold on
plot(yhatk)
hold off
%axis([time(1) time(end) 0.8*min(data(10:end)) 1.2*max(data(10:end)) ])
legend('True data', sprintf('%i-step predition', k), 'Validation starts', 'Location','NW')
title('US tobacco production')


%% How does the model work?
% We focus only on the validation data. Note that we now also remove the
% inital corrupted samples due to the filtering.
% Note that the prediction residual should only be white if k=1.
figure
acfEst = acf( ehat, noLags, 0.05, 1 );
title( sprintf('Prediction residual, %i-step prediction', k) )
fprintf('This is a %i-step prediction. Ideally, the residual should be an MA(%i) process.\n', k, k-1)
checkIfWhite( ehat );


%% What does the D'Agostino-Pearson's K2 test indicate?
% As the PACF should ring for an MA(k-1) process, we only check the ACF.
checkIfNormal( acfEst(k+1:end), 'ACF' );