/little_mcmc

Module for finding the argmax of any given function via a bared MCMC algorithm.

Primary LanguageJupyter NotebookMIT LicenseMIT

little_mcmc

Description

This is a little module of bared MCMC algorithm, used for sampling, or then computing the maximum value of any given function on any given domain of any given dimension.

Reference

Metropolis sampler:

  1. An Introduction to MCMC for Machine Learning, section 3.1.

  2. Metropolis et al (1953).

Global argmax:

  1. An Introduction to MCMC for Machine Learning, section 3.2.

To-do

[x] Gibbs sampling.

[ ] Try to finish the proof of Metropolis algorithm, clarifying every the reason of requiring those conditions in Metropolis algorithm.

[ ] Read the book: Bayesian Learning for Neural Networks.