Why does it take about 200 times longer to run the case code on a personal computer than the time shown on the web case?

1

(A Primer on Bayesian Methods for Multilevel Modeling,Example: Radon contamination Gelman and Hill [2006])

with pooled_model:
pooled_trace = pm.sample(random_seed=RANDOM_SEED)

web example : Sampling 4 chains for 1_000 tune and 1_000 draw iterations (4_000 + 4_000 draws total) took 4 seconds.

“actually”:Sampling 4 chains for 1_000 tune and 1_000 draw iterations (4_000 + 4_000 draws total) took 728 seconds.

WIN 10, RAM :16.0 GB ,
processor： AMD Ryzen 7 4800H with Radeon Graphics 2.90 GHz

2

Try updating to the very last version of PyMC, and if it’s still slow setting chains=1 in pm.sample. Does that make a big difference?

3

Thanks for the comments.
Followed your suggestion, updated Pymc, and set chains =1, but still no change in sampling time.
Do you have any other possible suggestions?

4

Are you getting any warning when you import PyMC?

What does the following show?

import pymc
print(pymc.__version__)
import pytensor
print(pytensor.__version__)
print(pytensor.config.config.blas__ldflags)
1 Like
5

Thanks for your help, there were some warnings I ignored.
After installing “m2w64-toolchain”, the sampling calculation speed is normal.

4 Likes
6

Glad you fixed your issue. As always, it it recommended to following the installation procedures found here (edit: …which should ensure that you have all relevant dependencies installed alongside PyMC).

1 Like