Open YSanchezAraujo opened 1 year ago
looking at advance function in pgas
it looks like it's not possible in this formulation? In my case:
rand(init_step(model)) #
# will just give a random element of:
(
truncated(Normal(m.theta.lam_lapse_init, 0.1), lower=-10),
truncated(Normal(m.theta.sigma_set_init[1], 0.1), lower=0.),
truncated(Normal(m.theta.sigma_set_init[2], 0.1), lower=0.),
truncated(Normal(m.theta.sigma_set_init[3], 0.1), lower=0.),
MvNormal(m.theta.mu_init, 1.)
)
it seems the workout would be to allow for
rand.(init_step(model))
?
cc @FredericWantiez
I'm wondering if a model like the one I present below is possible? The basic problem is one where the state isn't a single distribution, but a collection of distributions, which all evolve in a Markovian manner. I don't know exactly how this works internally, so the code is based on the assumption that the state is propagated forward from initialization to transition to observation.