Open
Description
Describe the issue:
I recently upgraded PyMC from version 5.6.1 to 5.16.2. After this upgrade, a specific model I use began freezing on a Linux cluster when processing certain datasets. However, when I run the same model on Windows, it completes without issues.
I'm wondering if there are any necessary code adaptations due to the version change or if this could be a platform-specific issue with the new version of PyMC.
Reproduceable code example:
import pymc as pm
with pm.Model() as model:
Xobs_rain_bg = pm.MutableData('features', X, dims='id1')
r = pm.Gamma("r", alpha=1, beta=0.1)
beta_bg_rain = pm.HalfNormal("beta_bg_rain", sigma=0.5)
mu = pm.Normal("mu", mu=0, sigma=10)
# Calculation of beta
beta = pm.math.exp(-(mu + beta_bg_rain * Xobs_rain_bg) / r)
# Modeling
y_obs = pm.Weibull("y_obs", alpha=r, beta=beta, observed=y, dims='id1')
trace = pm.sample(draws=1000, tune=1000, cores=4, random_seed=123)Error message:
No response
PyMC version information:
PyMC version: 5.16.2 (previously 5.6.1)
Python version: 3.11
Operating System: Linux Cluster (Issue occurs) / Windows (No issues)
Context for the issue:
No response