I am doing Bayesian inference of a few parameters (~3), albeit of a highly nonlinear model, and I am interested in the performance of random-walk Metropolis-Hastings versus Hamiltonian Monte Carlo methods. I don’t expect much from MH, but the traces seem particularly bad (below for one of the parame…

[image] caesoma: if not implementation from scratch is not difficult, though I’d prefer to be able to use something out of the box. You might consider submitting a pull request to AdvancedMH.jl to add a tuning option.

Tune Metropolis-Hastings (normal distribution) scale parameters (and algorithms parameters in general)

General Usage Performance

caesoma October 13, 2023, 5:32pm 11

That sounds like a great idea.

2 Likes

Adaptive proposals in AdvancedMH.jl (or other Metropolis-capable) package

Topic		Replies	Views
Access acceptance rate parameter for AdvancedMH in Turing Probabilistic Programming question , turing	5	149	September 16, 2024
Improving metropolis-hastings code General Usage monte-carlo	21	4796	September 6, 2021
Adaptive proposals in AdvancedMH.jl (or other Metropolis-capable) package Performance mcmc	0	212	February 27, 2024
Default sampler for mixed continuous/discrete parameters in Turing.jl Probabilistic Programming	2	118	October 15, 2025
Different samples distribution after execution Probabilistic Programming question , turing	5	648	September 24, 2020

Tune Metropolis-Hastings (normal distribution) scale parameters (and algorithms parameters in general)

Related topics