Pm.sample gets stuck after init with cores > 1

[mishooax]

@lucianopaz

ditching pythons multiprocessing module and running several sample jobs concurrently with different initialization seeds

You mean something like:

trace_1 = pm.sample(draws=2000, chains=1, cores=1, tune=2000, random_seed=SEED1)
trace_2 = pm.sample(draws=2000, chains=1, cores=1, tune=2000, random_seed=SEED2)

and run these in parallel somehow? (how?)

or two completely separate python processes, each sampling one MCMC chain + post-processing (combine the chains) when both are done? I guess this setup can be parallelized with something like mpi4py