Microcanonical Monte Carlo

Microcanonical Hamiltonian Monte Carlo (MCHMC) and its cousin Microcanonical Langevin Monte Carlo (MCLMC) constitute a new sampling algorithm for distributions with differentiable log likelihoods, introduced with the goal of replacing the current state of the art differentiable sampler, NUTS.

A Python implementation is available in Blackjax: https://blackjax-devs.github.io/sampling-book/algorithms/mclmc.html

Overview

Contents of this website

This website exists as a supplement to those papers, in order to explain the theory behind the algorithm in more detail, and document a variety of applications to which it is presently being applied.