https://github.com/mrquincle/cool-walking

Comparison different MCMC methods
https://github.com/mrquincle/cool-walking

Last synced: 3 months ago
JSON representation

Comparison different MCMC methods

• Host: GitHub
• URL: https://github.com/mrquincle/cool-walking
• Owner: mrquincle
• Created: 2017-03-16T00:44:25.000Z (over 8 years ago)
• Default Branch: master
• Last Pushed: 2017-04-05T14:36:49.000Z (over 8 years ago)
• Last Synced: 2025-03-15T20:42:12.510Z (7 months ago)
• Language: Matlab
• Size: 86.9 KB
• Stars: 0
• Watchers: 2
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

          
# Comparison different MCMC methods

Cool Walking and methods like J-Walking, S-Walking, Smart Darting, and Parallel Tempering are compared in [1]. In this repository we do not a full comparison (yet), but compare it with a new MCMC approach in which we do not used chains on different temperatures, but pick from a set of (symmetric) proposal distributions.

## Potential function

The potential function at different temperatures. 

![Potential function](https://github.com/mrquincle/cool-walking/blob/master/pictures/potential_energy.png?raw=true "Potential function")

For `T=0.1` we have only a few peaks. For `T=3.0` we have a much more uniform distribution.

The warmer distribution is used to initiate larger jumps.

## Results

The current results with adding complexity in the proposal distribution rather than smoothing the target distribution:

![MCMC hybrid](https://github.com/mrquincle/cool-walking/blob/master/pictures/mcmc_hybrids.png?raw=true "MCMC hybrid")

# References

[1] Cool Walking: A New Markov Chain Monte Carlo Sampling Method (Brown, Head-Gordon, 2002)