I’m having problems with sampling my model, the sampling simply won’t start and the progress bar is stuck at 0%. I’ve tried using both NUTS and Metropolis and neither works. The find_MAP method works fine and the MAP parameters are sensible. I’d like to know how to debug these types of problems, at…

Indeed it is. It works fine with cores=1. Do you have any idea how to fix this? The difference between this model and others that I’ve sampled successfully with multiple chains is minimal, not sure why it’s suddenly failing.

Hmmm so it works on other models? That could be a memory issue.

Are you thinking that the call to pickle, which we use for the multiprocessing, is what does it? I’m honestly not sure whether the observed data is replicated, and might cause such a problem.

Memory usage seems fine, i have 32GB of RAM on my machine. Here are the two models, the marginalized version fails with multiple cores. class PointSourcePointLens(pm.Model): def __init__(self, data, use_joint_prior=True, name='', model=None): super(PointSourcePointLens, self).__ini…

There’s a chance the problem is with the lambda's in some of those methods, though I’d expect an Exception, rather than just hanging. As an aside, subclassing Model is very cool!

Debugging models when sampling is stuck at 0%

Questions

junpenglao February 19, 2019, 5:26pm 2

Seems like it is a multi-chain sampling problem - try doing trace = pm.sample(..., cores=1)

TypeError: Random() got an unexpected keyword argument 'point' when running sample_posterior_predictive

Topic		Replies	Views
Sampling doesn't start when njobs > 1 for some models Questions	20	4895	January 17, 2020
Frequently Asked Questions Questions	12	25896	June 30, 2023
Shape error after pm.math.dot in multivar logistic regression model Questions	1	692	September 12, 2019
New User - NUTS Running forever on regression tutorial	8	238	February 14, 2025
Runtime Exception with NUTS v3	16	1189	March 16, 2022

Debugging models when sampling is stuck at 0%

Related topics