- #1
celestra
- 18
- 0
[SOLVED] Markov chain problem
Hi, all! This is not a homework thing or something like that. I'm just curious.
Suppose a Markov chain p(n+1)=Q*p(n), where Q=[q11, q12; q21, q22] is the transition probability matrix and p(n)=[p1(n);p2(n)] is the probability vector.
Given Q and p(1), we can generate a realization from it, i.e., a sequence of ones and zeros.
Now I want to do the inverse. Given a long enough sequence of ones and zeros, Q can be obtained by counting the number of changes from zero to zero(which will be q11), zero to one(q12), one to zero(q21), and one to one(q22). But, how can I get the initial probability vector p(1) from this sequence? Maximum Likelihood Estimation, blah blah? I don't know about that. Please, someone give me a clue how I can do this. Or is it impossible to do that from just only one realization sequence?
Thanks in advance.
Hi, all! This is not a homework thing or something like that. I'm just curious.
Suppose a Markov chain p(n+1)=Q*p(n), where Q=[q11, q12; q21, q22] is the transition probability matrix and p(n)=[p1(n);p2(n)] is the probability vector.
Given Q and p(1), we can generate a realization from it, i.e., a sequence of ones and zeros.
Now I want to do the inverse. Given a long enough sequence of ones and zeros, Q can be obtained by counting the number of changes from zero to zero(which will be q11), zero to one(q12), one to zero(q21), and one to one(q22). But, how can I get the initial probability vector p(1) from this sequence? Maximum Likelihood Estimation, blah blah? I don't know about that. Please, someone give me a clue how I can do this. Or is it impossible to do that from just only one realization sequence?
Thanks in advance.