Hello, I am currently implementing a hierarchical linear model (with correlated slopes and intercepts) that includes various random effects. I am using a single predictor. The dataset is huge, nearly 300k samples. When I try to sample (4 chains, 1000 samples, 1000 tuning iterations), the sampler ta…

It sounds like the dataset size is the simplest explanation. You might try sampling with sample_numpyro_nuts, particularly if you can get your hands on a GPU (e.g. on Colab). I get significant speed gains for my larger models. Another solution is to use ADVI, for which your milage may vary in terms …

Really long sampling time

Questions v3

fonnesbeck April 22, 2022, 3:07pm 5

Yes, I strongly recommend running the v4 beta:

pip install -U pymc --pre

I’ve had no issues with sample_numpyro_nuts in v4. so hopefully this will fork for you.

Topic		Replies	Views
Hierarchical Model - Slow Sampling Questions	4	1191	March 26, 2020
PyMC3 slows rapidly with increasing numbers of parameters Questions	5	11144	December 3, 2018
Example Code Taking Very Long time to run? Questions	2	1520	November 12, 2020
Pymc3 JAX: negligible speedup Development	4	2328	January 8, 2021
PyMC v4 sampling efficiency questions v5	9	1751	July 11, 2022

Really long sampling time

Related topics