I’m working on a embedding model that seems to suffer from a large number of isolated local maxima. I’m currently trying to recover the original coordinates of a dataset generated synthetically, but chains initialized randomly get stuck in local maxima when compared to a chain initialized at the gro…

From your description it sounds like you might be talking about the tempering during a trajectory approach described by Radford Neal in Section 5.7 of his MCMC using Hamiltonian Dynamics chapter in the Handbook of Markov chain Monte Carlo? If so I think you are correct that it would problematic usin…

matt: very nice summary! just to echo matt on “adaptive path sampling”: it uses nuts x tempering with a continuous temperature. The linked GitHub link has a semi-automated implantation, which basically runs an augmented stan file iteratively. I would also add another practical solution is stacking…

Potentially also relevant: [image] Chainsail, the web service for sampling multimodal probability distributions, is now open-source! Publicity In August '22, I presented Chainsail , an internal project my colleagues and me over at Tweag have been working on for quite a whil…

Thanks to all of you, this gives me a lot of approaches to try!

Handling multimodality in HMC/Stan - tempering & NUTS?

Modeling

jsocolar June 24, 2023, 10:22pm 4

Potentially also relevant:

Topic		Replies	Views
Simulated tempering and multimodal sampling Algorithms	1	1157	September 2, 2020
Multi-modality of posteriors Modeling fitting-issues	3	909	December 10, 2018
Recommendations for problem-specific Stan-esque dynamic HMC implementation General algorithms	2	766	April 13, 2022
Momentum Diffusion HMC? Algorithms	40	5202	August 24, 2017
Using samples from the adaptation / warmup phase in NUTS General	8	1431	December 21, 2022

Handling multimodality in HMC/Stan - tempering & NUTS?

Related topics