Bayesian Optimization for Probabilistic Programs

Authors: Tom Rainforth, Tuan Anh Le, Jan-Willem van de Meent, Michael A. Osborne, Frank Wood

NeurIPS 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We present applications of our method to a number of tasks including engineering design and parameter optimization. We first demonstrate the ability of BOPP to carry out unbounded optimization using a 1D problem with a significant prior-posterior mismatch as shown in Figure 4. Next we compare BOPP to the prominent BO packages SMAC [14], Spearmint [25] and TPE [3] on a number of classical benchmarks as shown in Figure 5.
Researcher Affiliation Academia Department of Engineering Science, University of Oxford College of Computer and Information Science, Northeastern University
Pseudocode No The paper includes a high-level algorithm overview in Figure 3 but does not provide formal pseudocode blocks or labeled algorithms.
Open Source Code Yes Code available at http://www.github.com/probprog/bopp/ Code available at http://www.github.com/probprog/deodorant/
Open Datasets No The paper uses benchmark problems and simulations (e.g., 'Energy2D simulations', 'Hartmann 6D', 'SVM on-grid', 'LDA on-grid', 'pickover attractor') rather than traditional public datasets with explicit access information.
Dataset Splits No The paper evaluates on benchmark functions and simulations, not traditional datasets with specified train/validation/test splits.
Hardware Specification No The paper does not provide specific details about the hardware (e.g., CPU, GPU models) used for running the experiments.
Software Dependencies No The paper mentions software like Anglican, Energy2D, Stan, Church, Venture, Web PPL, but does not provide specific version numbers for these or other software dependencies used in their experimental setup.
Experiment Setup Yes BOPP therefore employs an affine scaling to a [ 1, 1] hypercube for both the inputs and outputs of the GP. We use as a default covariance function a combination of a Mat ern3/2 and Mat ern-5/2 kernel. Inference over hyperparameters is performed using Hamiltonian Monte Carlo (HMC) [6]. r is a parameter set to 1.5re by default. ...using a variant of annealed importance sampling [19] in which lightweight Metropolis Hastings (LMH) [28] with local random-walk moves is used as the base transition kernel.