使用带有 depmixS4 包的 R 的隐马尔可夫模型（HMM）

Question

I am using depmixS4 package to use HMM.

After using posterior function, I get 2 columns. First column with currently which state I may be in(seems like using Viterbi Algorithm), and the next column is the probability of for being in each state.

I think that the first column with the state should match with the maximum probability in second column, but the result shows that that is not always the case. For example if the probability for state 1 : 0.2, state 2 : 0.7, state 3 : 0.1, the first column that shows which state I am in may show that I am in state 1. (intuitively should show that I am in state 2)

https://cran.r-project.org/web/packages/depmixS4/depmixS4.pdf on here, it says that the second column shows "delta prob", but I cannot not find that word even in the reference.

Thanks for reading

Answer 1

I'm not sure the above answer is correct. The state is a combination of the previous states and the probability of the current observed value occurring under the 3 distributions. To see the code that generates the deltas take a look at

depmixS4::viterbi

Where you can see the states generated in the "backwards" part of the viterbi algorithm

Answer 2

The delta probabilities are defined in Rabiner(1989) at equation 30.

The delta probabilities give you a vector of probabilities for each state. Then you use this vector to simulate the state; this means that you will simulate state 1, 20% of the time, state 2, 70% and so on. I am guessing that the package gives you one simulation. Hope that helps.