NUTS doesn't run even with the simplest of models?

Questions
1

I am just getting started with PyMC3 and have tried to run a extremely simple example from Osvaldo Martin’s book, but NUTS seems to get hung up during initialization:

data = stats.bernoulli.rvs(p=0.35, size=4)
with pm.Model() as simple_model:
theta = pm.Beta(‘theta’, alpha=1., beta=1.)
y = pm.Bernoulli(‘y’, p=theta, observed=data)
trace = pm.sample(1000, random_seed=123)

gives:

Auto-assigning NUTS sampler…
Initializing NUTS using jitter+adapt_diag…
Multiprocess sampling (4 chains in 4 jobs)
NUTS: [theta]

If I set chains=1 it initializes with 1 job and runs. However, if I set chains=2 and cores=2, it hangs again. What might be going wrong? I am using Python 3.8.3 and Anaconda 2020.11 on Windows 10.

2

Unfortunately parallelism on windows is still a bit unstable sometimes. The next version 3.10 which is coming out soon should have some fixes for that, however.

3

Thanks for the reply. I suppose that means there isn’t a short-term work around for running parallel jobs? In any case, looking forward to your next release then!

4

Some people have reported good success using WSL2

5

3.10 is out now. Can you try if that fixes your parallelism problem?