camir-aes2014: toolboxes/FullBNT-1.0.7/Kalman/learn

first hg version after svn

rev	line source
wolffd@0	1 function [coef, C] = learn_AR(data, k)
wolffd@0	2 % Find the ML parameters of a vector autoregressive process of order k.
wolffd@0	3 % [coef, C] = learn_AR(k, data)
wolffd@0	4 % data{l}(:,t) = the observations at time t in sequence l
wolffd@0	5
wolffd@0	6 warning('learn_AR seems to be broken');
wolffd@0	7
wolffd@0	8 nex = length(data);
wolffd@0	9 obs = cell(1, nex);
wolffd@0	10 for l=1:nex
wolffd@0	11 obs{l} = convert_to_lagged_form(data{l}, k);
wolffd@0	12 end
wolffd@0	13
wolffd@0	14 % The initial parameter values don't matter, since this is a perfectly observable problem.
wolffd@0	15 % However, the size of F must be set correctly.
wolffd@0	16 y = data{1};
wolffd@0	17 [s T] = size(y);
wolffd@0	18 coef = rand(s,s,k);
wolffd@0	19 C = rand_psd(s);
wolffd@0	20 [F,H,Q,R,initx,initV] = AR_to_SS(coef, C, y);
wolffd@0	21
wolffd@0	22 max_iter = 1;
wolffd@0	23 fully_observed = 1;
wolffd@0	24 diagQ = 0;
wolffd@0	25 diagR = 0;
wolffd@0	26 [F, H, Q, R, initx, initV, loglik] = ...
wolffd@0	27 learn_kalman(obs, F, H, Q, R, initx, initV, max_iter, diagQ, diagR, fully_observed);
wolffd@0	28
wolffd@0	29 [coef, C] = SS_to_AR(F, Q, k);
wolffd@0	30

Mercurial > hg > camir-aes2014