I am trying to add a hierarchy on a constrained weight, but the centered parameterization is tripping me up. I think I have the non-centered version, but would prefer to test the centered version first as that is best practice. # w number of weights, g number of groups ## non-centered sigma_w = pm.…

Neither centered nor non-centered is “best practice”, there are just cases where you want to use one instead of the other. In general, non-centered is better when the sigma parameter admits quite small values (the groups are very similar), because this makes “funnels” that are hard to sample, even f…

Hierarchy on Half Normal Distribution

Questions v5

ricardoV94 July 10, 2024, 6:39am 8

That’s why I mentioned lognormal, which has that implicit exp baked in

3 Likes

Topic		Replies	Views
Parametrize mixed beta regression in non centered way Questions	6	1194	April 17, 2018
Improving speed with large dimensional problem Questions	12	4080	September 27, 2017
Hierarchical model with bounded Normals? Questions	3	590	January 20, 2020
Ideas for reparameterizing models/changing priors to avoid divergences Questions	3	956	December 5, 2019
Non-centered hierarchical model produces weird looking posteriors version agnostic modeling	2	462	November 28, 2023

Hierarchy on Half Normal Distribution

Related topics