1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
|
module Probability.Distribution.Markov where
import Probability.Random
sample_markov x0 next = do
x1 <- next x0
xs <- sample_markov x1 next
return (x0:xs)
sample_markov_n 0 x0 next = return []
sample_markov_n n x0 next = do
x1 <- next x0
xs <- sample_markov_n (n-1) x1 next
return (x0:xs)
-- I think if next has type (a->Dist a) and not just
-- (a->Sample a), then we could compute likelihoods.
-- The fastest way would be to represent the transition
-- probabilities in a matrix -- at least for integer
-- sequences -- but that is less general. Maybe we should
-- make a markov_int distribution that takes a starting
-- distribution and a transition probability matrix?
-- I presume that we would need to make any markov DISTRIBUTION
-- have an actual length. Do we want them to be lazy? Strict?
|
