Nearly Linear Row Sampling Algorithm for Quantile Regression

Authors: Yi Li, Ruosong Wang, Lin Yang, Hanrui Zhang

ICML 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we conduct an empirical evaluation of our sampling scheme against the three algorithms proposed by Yang et al. (2013), which are the current state of the art.
Researcher Affiliation Academia 1School of Physical and Mathematical Sciences, Nanyang Technological University, Singapore 2Department of Computer Science, Carnegie Mellon University, USA 3Department of Electrical and Computer Engineering, University of California Los Angeles, USA 4Department of Computer Science, Duke University, USA.
Pseudocode Yes Algorithm 1 Algorithm for solving quantile regression
Open Source Code No The paper does not include an unambiguous statement that the authors are releasing the code for the work described in this paper, nor does it provide a direct link to a source-code repository.
Open Datasets Yes Census Data. We also evaluate our algorithm on a real-world dataset, the U.S. 2000 Census data, which consists of salary and related information on people who met certain criteria. As in (Yang et al., 2013), we conduct an experiment on the same 5% sample of the census data. [...] http://www.census.gov/census2000/PUMS5.html
Dataset Splits No The paper does not specify exact split percentages or absolute sample counts for training, validation, and testing, nor does it reference predefined splits with citations.
Hardware Specification Yes All tests are run under MATLAB 2019b on a machine of Intel Core i7-6550U CPU@2.20GHz with 2 cores.
Software Dependencies Yes All tests are run under MATLAB 2019b on a machine of Intel Core i7-6550U CPU@2.20GHz with 2 cores.
Experiment Setup Yes In a similar manner to (Yang et al., 2013), rather than determining the sample size from a given distortion parameter ε, we vary the quantile parameter τ {0.5, 0.75, 0.95}, and for each τ, vary the sample size among {100, 200, . . . , 1000}. Here the parameter τ corresponds to the parameter in the function hτ(t) defined in Equation (1) rather than the renormalized function ρτ(t) defined in Equation (3). We choose n = 105 and d = 50.