Calibrated Approximate Bayesian Inference

Authors: Hanwen Xing, Geoff Nicholls, Jeong Lee

ICML 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We apply the algorithms to calibrate a pseudo-likelihood approximation. In Section 5, we calibrate the approximate posterior of a random partition in a hierarchical model with a Dirichlet Process prior on the distribution of random effects.
Researcher Affiliation Academia 1Department of Statistics, University of Oxford, UK 2Department of Statistics, University of Auckland, New Zealand.
Pseudocode Yes Algorithm 1 Estimation of operational coverage c(y)... Algorithm 2 AIS Estimation of operational coverage c(y)... Algorithm 3 Estimation of operational coverage c(y) via regression
Open Source Code No The paper does not provide concrete access to source code (e.g., a specific repository link or an explicit code release statement) for the methodology described.
Open Datasets Yes Figure 1 is a 200 by 200 binary image obtained by thresholding a grey-level image of ice floes from Banfield & Raftery (1992). ... Our dataset has the classical format of a complete design, with five categorical variables, including Treatment (N = 12 levels), and four block variables, B1 (with q = 3 levels) B2 (r = 2 levels), B3 (two levels) and B4 (seven levels) so that we have n = 1008 observations, y = (y1, . . . , yn).
Dataset Splits Yes We simulate M = 810 pairs {c(i), y(i)}M i=1 of training data following Algorithm 3. ... In order to further test the robustness of our method, we partition the full synthetic dataset {c(i), y(i)}M i=1 into four equal-size subsets and fit a Probit BART model using formula M2 to each subset.
Hardware Specification No The paper does not provide specific hardware details (e.g., exact GPU/CPU models, processor types with speeds, memory amounts, or detailed computer specifications) used for running its experiments.
Software Dependencies No The paper mentions the R package "bart Machine" and "lme4" but does not specify their version numbers, nor does it list versions for other ancillary software or libraries.
Experiment Setup Yes We initialise the AIS sampler with M = 1000 samples {phi_i, y_i}M i=1 with phi_i pi(phi|yobs) and y_i p_F(y_i|phi_i), the true likelihood. We set the number of AIS iterations NAIS = 60 with cooling schedule beta_j = 1.05j for j = 1, . . . , NAIS, gamma_j = 0.02j for j = 1, . . . 50 and gamma_j = 1 for j = 51, . . . , NAIS at the jth iteration.