search

davharris / mistnet

stochastic neural networks in R
24 stars 3 forks source link

Annealed Importance Sampling #69

Open davharris opened 10 years ago

davharris commented 10 years ago

Depends on #67 and #99

davharris commented 10 years ago

Q should have the same marginals as P but all its elements should be independent. Can get that by doing 1000 steps of regular importance sampling (i.e. sampling from the prior and weighing by the likelihood), and estimating the probabilities associated with each species.

Start the chain at a random sample from Q.

davharris commented 10 years ago

Do I need the annealing? Why not just sample from the marginal distribution estimated by mean field and see how that does?