Open BioTurboNick opened 2 years ago
Hi,
Here is an example with the Weather guessing game from Wikipedia.
Does that makes things clearer?
# K: number of states
# a[i] = probability of starting in i
# vector of length K
start_probability = [0.6, 0.4]
# A[i,j] = probability of going from i to j
# matrix of size KxK
transition_probability = [
0.7 0.3; # i=1 (Rainy)
0.4 0.6 # i=2 (Sunny)
]
# B[i] = probability distribution of the observations in state i
# vector of size K
# We can model an emission matrix with the discrete `Categorical` distribution:
emission_probability = [
# [walk, shop, clean]
Categorical([0.1, 0.4, 0.5]), # i=1 (Rainy)
Categorical([0.6, 0.3, 0.1]), # i=2 (Sunny)
]
hmm = HMM(start_probability, transition_probability, emission_probability)
I'm coming in relatively fresh to HMMs and I'm having trouble matching up terms I'm seeing in other work.
1) "Emission matrix" doesn't appear in the documentation - I gather it can be provided in place of
B
, but the documentation doesn't show that or describe its form.2) The form the "Transition matrix" should take isn't explicitly documented.
I'd be happy to make a contribution here, once I understand them.