Hamiltonian Annealed Importance Sampling (HAIS)

HAIS is a method for combining Hamiltonian Monte Carlo (HMC) and Annealed Importance Sampling (AIS), so that a single Hamiltonian trajectory can stretch over many AIS intermediate distributions. This greatly improves the efficiency of AIS in continuous state spaces. It is described in detail in the paper:

J Sohl-Dickstein, BJ Culpepper
Hamiltonian annealed importance sampling for partition function estimation
Redwood Technical Report (2011)
http://arxiv.org/abs/1205.1925

The code in this repository can be used for log likelihood estimation, partition function estimation, and importance weight estimation. It can also be used as a Hamiltonian Monte Carlo sampler. See HAIS_examples.m for usage examples.

Files

• HAIS_examples.m demonstrates the capabilities of this code in a variety of scenarios.
• HAIS.m performs Hamiltonian Annealed Importance Sampling.
• HAIS_logL.m calculates the log likelihood of a model given data using HAIS.
• HAIS_logL_aux.m calculates the log likelihood of a model with hidden (auxiliary) variables given data using HAIS.