The Falling Factorial Basis and Its Statistical Applications

Authors: Yu-Xiang Wang, Alex Smola, Ryan Tibshirani

ICML 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We provide no theory but demonstrate excellent empirical results, improving on, e.g., the maximum mean discrepancy (Gretton et al., 2012) and Anderson-Darling (Anderson & Darling, 1954) tests. and 5.2. Numerical experiments We examine the higher order KS tests by simulation.
Researcher Affiliation Academia Yu-Xiang Wang YUXIANGW@CS.CMU.EDU Alex Smola ALEX@SMOLA.ORG Ryan J. Tibshirani RYANTIBS@STAT.CMU.EDU Carnegie Mellon University, 5000 Forbes Ave, Pittsburgh, PA 15213
Pseudocode Yes Algorithm 1 Multiplication by H(k) and Algorithm 2 Multiplication by (H(k)) 1
Open Source Code No The paper mentions that their implementation of falling factorial transforms uses MEX functions for comparison with highly-optimized libraries, implying custom code, but does not explicitly state that this code is open-sourced or provide a link for it. It only states the paper and supplement are on arXiv.
Open Datasets No The experiments use synthetically generated data drawn from standard statistical distributions (Normal, t-distribution, Laplace) rather than pre-existing publicly available datasets with specific access information like links or citations.
Dataset Splits No The paper describes generating synthetic data for numerical experiments but does not provide specific dataset split information (e.g., percentages or counts for train/validation/test sets), cross-validation setup, or citations to predefined splits.
Hardware Specification No The paper states: 'The experiments were performed on a laptop computer.' This is not a specific hardware detail like a GPU/CPU model or memory amount.
Software Dependencies No The paper mentions 'Matlab' and 'Stanford Wave Lab' functions, but does not provide specific version numbers for these software components or any other key dependencies.
Experiment Setup Yes The paper states: 'Figures 3 and 4 show the results of two experiments in which n = 100 and R = 1000.' It also describes the setup for numerical experiments: 'The setup: we fix two distributions P, Q. We draw n i.i.d. samples X(n), Y(n) P, calculate a test statistic, and repeat this R/2 times; we also draw n i.i.d. samples X(n) P, Y(n) Q, calculate a test statistic, and repeat R/2 times.'