For simulations to determine convergence rates of gibbs_sampler, I've been setting burnins=0, delay=1, draws_per_restart=X where X is the number of passes I want to make. I think this should be the default way we call gibbs_sampler, by just giving it a max number of passes. So, it'd look like:
def gibbs_sampler(cp, max_passes):
Then, an API user can take the returned data and use any parameters they like for the delay, burnins, etc. This removes draws/restart as a parameter, but that could be replicated by just repeating the call.
For simulations to determine convergence rates of
gibbs_sampler
, I've been settingburnins=0
,delay=1
,draws_per_restart=X
whereX
is the number of passes I want to make. I think this should be the default way we callgibbs_sampler
, by just giving it a max number of passes. So, it'd look like:Then, an API user can take the returned data and use any parameters they like for the delay, burnins, etc. This removes draws/restart as a parameter, but that could be replicated by just repeating the call.