PyMC3 produces weird results for this multiple linear regression model. I tried the same model with Stan and Numpyro. The results of both are quite reasonable for the same model. The distribution parameters seem to be same across the three systems. I tried all the different initialization strategies…

[image] rusty: pm.StudentT('w',78.04098510742188,-17.60015869140625,68.0557632446289, shape=(10,)) Can you try with pm.StudentT('w',78.04098510742188,-17.60015869140625,sigma=68.0557632446289, shape=(10,))

Pymc3 produces different results than Stan/NumPyro

Questions v3

twiecki March 15, 2022, 10:57am 4

Does the sampler give any indication that it had problems sampling? The results look really wrong.

Topic		Replies	Views
Converting Model from Stan Questions	15	3176	June 27, 2017
When ever I tried running NUTS sampling my notebook gets disconnected from kernel Questions	4	1872	July 7, 2018
Runtime Exception with NUTS v3	16	1138	March 16, 2022
Sample method of pymc3 Questions	4	685	July 15, 2018
The example doesn't converge well and posteriors fluctuated Questions	3	584	March 21, 2019

Pymc3 produces different results than Stan/NumPyro

Related topics