/l2hmc

TensorFlow implementation for training MCMC samplers from the paper: Generalizing Hamiltonian Monte Carlo with Neural Network

Primary LanguageJupyter NotebookApache License 2.0Apache-2.0

L2HMC: Automatic Training of MCMC Samplers

TensorFlow open source implementation for training MCMC samplers from the paper:

Generalizing Hamiltonian Monte Carlo with Neural Networks

by Daniel Levy, Matt D. Hoffman and Jascha Sohl-Dickstein


Given an analytically described distributions (implemented as in utils/distributions.py), L2HMC enables training of fast-mixing samplers. We provide an example, in the case of the Strongly-Correlated Gaussian, in the notebook SCGExperiment.ipynb --other details are included in the paper.

Contact

Code author: Daniel Levy

Pull requests and issues: @daniellevy

Citation

If you use this code, please cite our paper:

@article{levy2017generalizing,
  title={Generalizing Hamiltonian Monte Carlo with Neural Networks},
  author={Levy, Daniel and Hoffman, Matthew D. and Sohl-Dickstein, Jascha},
  journal={International Conference on Learning Representations},
  year={2018}
}

Note

This is not an official Google product.