Markov Chain - Hidden Markov Model; how to create markov chains and combine them to a hidden markov model
6 vues (au cours des 30 derniers jours)
Afficher commentaires plus anciens
i have a problem from getting from an markov chain to a hidden markov model. I want to model the activity of a person, which is given in 15 minute intervals. First of all i want to create a markov chain for a single day where i have (no activity or activity in this intervals). Furthermore i have a month of data, so every additional day will be a markov chain in parallel. So my question is how to create markov chains with the implemented functions and how to combine these chains to get a complete hidden markov model. How can the states be merged?
A day may look like this starting from 0 - 24h. activity is given in the time intervals as (activity = 1, no activity = 0):
first day : 0 0 0 0 0 1 1 1 0 0 1 0 0 0 0
second day: 0 0 0 1 0 1 1 1 0 0 1 0 1 0 0
...
So i want to create these two sequences and want to combine them. What about the initial state?
Thx
0 commentaires
Réponses (2)
Richard Willey
le 13 Juil 2011
When I am working with Markov Chains I'm normally looking at stationary distributions which, by definition, don't depend on the initial state. Alternatively, if I'm looking at a set of emissions, I typically do so after the burn in.
I'm attaching some code to show a simple use case.
%%Estimating the convergence probabilities for a Markov Chain
clear all
clc
% Assume a triple modular redundant system (three independent modules)
m = 3;
% At any given point in time, a module is either working or broken
% The probability that a module that is working at time T is broken at
% time T+1 is given by
pFail = .05; % Probability of a failure
% The probability that a module that is broken at time T is repaired at
% time T+1 is given by
pFix = .40; % Probability of a fix
% The system has 8 possible states
% (0,0,0) All 3 modules are working
% (0,0,1) Modules 2 and 3 are working, module 1 is broken
% (0,1,0) Modules 1 and 3 are working, module 2 is broken
% (0,1,1) Modules 1 and 2 are broken, module 3 is working
% (1,0,0) Modules 1 and 2 are working, module 3 is broken
% (1,0,1) Modules 1 and 3 are broken, module 2 is working
% (1,1,0) Modules 2 and 3 are broken, module 1 is working
% (1,1,1) All three modules are broken
% The transition matrix for the system is
n = 2^m; % Number of states
[x, y] = meshgrid(0:n-1, 0:n-1); % All possible combos of state1 and state2
state1 = dec2bin(x)-48; % Convert to binary
state2 = dec2bin(y)-48; % Convert to binary
P = pFail.^(state1 == 0 & state2 == 1) .* ... Subsystem fails
(1-pFail).^(state1 == 0 & state2 == 0) .* ... Subsystem does not fail
pFix.^(state1 == 1 & state2 == 0) .* ... Subsystem is fixed
(1-pFix).^(state1 == 1 & state2 == 1); % Subsystem is not fixed
% P contains individual probabilities for each subsystem. Multiply across
% the row to calculate total probability assuming subsystem events are
% independent. Then reshape and transpose to get right symmetric
P = reshape(prod(P,2), n, n);
P = P'
Simple illustrative example
% Let Pi = a vector of probabilities that describe the initial state of the
% system.
% Chose a random starting position for Pi at T = 0
Pi = 100 * rand(1, length(P));
Pi = Pi/sum(Pi)
% Multiple Pi * the transition matrix to get a vector of describing the
% expected state of the system at T = 1
Pi = Pi * P
% Repeat 19 more times to get the expected state of the system at time T =
% 20
for i = 1: 19
Pi = Pi * P;
end
% After 20 iterations, Pi will have converged very close to the steady
% state of the system
Pi
% Alternatively (and easier)
Pi_other = P^20;
Pi_other = Pi_other(1,:)
%%Use Eigenvectors to calculate the steady state
% The purpose of the original example was to illustrate that the steady
% state probability vector can be described using the equation
% Pi*P = Pi We can use this information to create a system of linear
% equations and solve for Pi
[V D] = eigs(P');
Steady_State = V(:,1)/sum(V(:,1))
1 commentaire
B
le 21 Juil 2011
Hey Richard,
really appreciated your work there. My mind got blown away and my tears dried, when i realised how awesome those lines are.
Thank You
Voir également
Catégories
En savoir plus sur Markov Chain Models dans Help Center et File Exchange
Produits
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!