- #1
Stochasticus
- 2
- 0
I'm trying to learn more about Markov chains and came across the Gibbs sampler
x_1{t+1} ~ p(x_1|x_2 = x_2{t},...x_n{t})
x_2{t+1} ~ p(x_2|x_1 = x_1{t+1},x_3 = x_3{t},...,x_n{t})
.
.
.
x_i{t+1} ~ p(x_i|x_1 = x_1{t+1},...,x_(i-1) = x_(i-1){t+1},x_(i+1) = x_(i+1){t},...,x_n{t})
Supposedly this thing is a Markov chain. I just don't see it. It sure looks like each updated variable x_i{t+1} is contingent on the whole set, not just the prior value x_i{t}. Can someone show me how this satisfies the Markov criterion
x_1{t+1} ~ p(x_1|x_2 = x_2{t},...x_n{t})
x_2{t+1} ~ p(x_2|x_1 = x_1{t+1},x_3 = x_3{t},...,x_n{t})
.
.
.
x_i{t+1} ~ p(x_i|x_1 = x_1{t+1},...,x_(i-1) = x_(i-1){t+1},x_(i+1) = x_(i+1){t},...,x_n{t})
Supposedly this thing is a Markov chain. I just don't see it. It sure looks like each updated variable x_i{t+1} is contingent on the whole set, not just the prior value x_i{t}. Can someone show me how this satisfies the Markov criterion