A Kernel Test of Goodness of Fit

Authors: Kacper Chwialkowski, Heiko Strathmann, Arthur Gretton

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

Reproducibility Variable Result LLM Response
Research Type Experimental We apply our test to quantifying convergence of approximate Markov Chain Monte Carlo methods, statistical model criticism, and evaluating quality of fit vs model complexity in nonparametric density estimation.Section 4 contains experimental illustrations on synthetic examples, statistical model criticism, bias-variance trade-offs in approximate MCMC, and convergence in nonparametric density estimation.
Researcher Affiliation Academia Gatsby Unit, University College London, United Kingdom
Pseudocode No The paper describes the test procedure in a list of steps in Section 2.2, but this is presented as plain text and not formatted as pseudocode or an algorithm block.
Open Source Code Yes Code can be found at https://github.com/karlnapf/kernel_goodness_of_fit.
Open Datasets No The paper mentions using the solar dataset but does not provide a direct link, DOI, or a citation with author names and year for public access.
Dataset Splits No The paper describes training and testing data splits (e.g., We fit Ntrain = 361 data..., ...remaining Ntest = 41 data) but does not explicitly mention a validation data split or its size.
Hardware Specification No The paper does not provide any specific details regarding the hardware used for experiments, such as GPU/CPU models, memory specifications, or cloud computing instance types.
Software Dependencies No The paper does not provide specific software dependencies with version numbers (e.g., PyTorch 1.9, Python 3.8, CUDA 11.1).
Experiment Setup Yes For a fixed number of degrees of freedom we drew 1400 samples and calculated the p-value. This procedure was repeated one hundred times...We emphasize the need for an appropriate choice of the wild bootstrap process parameter, an, which indicates the probability of a sign flip. In Figure 1 we plot p-values for an being set to 0.5. ... We have thinned the observations by a factor of 20 and set an = 0.1...Sample size was set to 500/1000, a_n=0.5.