HMM.zip_hidden markov_imageha5_partlyU46_wintersfn_马尔科夫代码

function [A_Matrix_New, Pi_Init_New, count] = Baum_Welch(LRR, T, State_N, A_Matrix_old, Pi_Init_old) % Baum_Welch Algorithm % 初始化HMM的转移概率矩阵和初始概率向量 A_Matrix_New = A_Matrix_old; Pi_Init_New = Pi_Init_old; eta = 1; % eta为转移矩阵与初始向量的各个参数平方和，为了进入循环迭代，初始设为1 count = 0; while eta > 0.01 % 当转移矩阵与初始向量的各个参数平方和小于0.01时收敛，迭代结束 % 更新参数 A_Matrix = A_Matrix_New; Pi_Init = Pi_Init_New; alpha = Forward_variable(LRR, A_Matrix, Pi_Init, T, State_N); beta = Backward_variable(LRR, A_Matrix, T, State_N); gamma = Gamma_variable(alpha, beta, T, State_N); ksi = Ksi_variable(LRR, alpha ,beta, A_Matrix, T, State_N); % 重新估计 reestimation Pi_Init_New = gamma(1, :); A_Matrix_New = zeros(State_N, State_N); for i = 1:State_N denominator = sum(gamma(1:(T-1), i)); for j = 1:State_N A_Matrix_New(i,j) = sum(ksi(:,i,j))/denominator; end end eta = sum(sum((A_Matrix_New - A_Matrix).^2)) + sum((Pi_Init_New - Pi_Init).^2); count = count+1; end % end of while end % end of function