Nonparametric Hamiltonian Monte Carlo

Authors: Carol Mak, Fabian Zaiser, Luke Ong

ICML 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental This paper introduces the Nonparametric Hamiltonian Monte Carlo (NP-HMC) algorithm which generalises HMC to nonparametric models. We provide a correctness proof of NP-HMC, and empirically demonstrate significant performance improvements over existing approaches on several nonparametric examples.
Researcher Affiliation Academia Carol Mak 1 Fabian Zaiser 1 Luke Ong 1 1Department of Computer Science, University of Oxford, United Kingdom. Correspondence to: Carol Mak <pui.mak@cs.ox.ac.uk>.
Pseudocode Yes Figure 4. Pseudocode for Nonparametric Hamiltonian Monte Carlo
Open Source Code Yes The code for our implementation and experiments is available at https://github.com/fzaiser/nonparametric-hmc and archived as (Zaiser & Mak, 2021).
Open Datasets Yes We used this model to generate N = 200 training data points for a fixed θ = (K = 9,µ 1...K ).
Dataset Splits Yes We used this model to generate N = 200 training data points for a fixed θ = (K = 9,µ 1...K ). We computed the log pointwise predictive density (LPPD) for a test set with N = 50 data points Y = {y1,...,y N }, generated from the same θ as the training data.
Hardware Specification No No specific hardware details (e.g., CPU/GPU models, memory) used for running experiments are provided in the paper.
Software Dependencies No We implemented the NP-HMC algorithm and its variants (NP-RHMC and NP-DHMC) in Python, using PyTorch (Paszke et al., 2019) for automatic differentiation. While PyTorch is mentioned, a specific version number is not provided, making it not reproducible based on the strict definition.
Experiment Setup Yes Table 1. Total variation distance from the ground truth for the geometric distribution, averaged over 10 runs. Each run: 10^3 NP-DHMC samples with 10^2 burn-in, 5 leapfrog steps of size 0.1; and 5 x 10^3 LMH, PGibbs and RMH samples.