Markovian Interference in Experiments

Authors: Vivek Farias, Andrew Li, Tianyi Peng, Andrew Zheng

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our empirical evaluation includes a set of experiments on a city-scale ride-hailing simulator.
Researcher Affiliation Academia Vivek Farias Sloan School of Management Massachusetts Institute of Technology Cambridge, MA 02139 vivekf@mit.edu Andrew A. Li Tepper School of Business Carnegie Mellon University Pittsburgh, PA 15213 aali1@cmu.edu Tianyi Peng Department of Aeronautics and Astronautics Massachusetts Institute of Technology Cambridge, MA 02139 tianyi@mit.edu Andrew Zheng Operations Research Center Massachusetts Institute of Technology Cambridge, MA 02139 atz@mit.edu
Pseudocode No No pseudocode or algorithm block is present in the provided text.
Open Source Code Yes Did you include the code, data, and instructions needed to reproduce the main experimental results (either in the supplemental material or as a URL)? [Yes]
Open Datasets Yes Riders and drivers are generated according to real world data, based on [1]; this yields 300k requests and 7k unique drivers per real day. [1] TLC Trip Record Data TLC. https://www1.nyc.gov/site/tlc/about/tlc-trip-recorddata.page.
Dataset Splits No The paper describes running experiments over simulated trajectories but does not specify explicit train/validation/test splits for a fixed dataset. The data for experiments is generated dynamically.
Hardware Specification No The paper states "See Appendix." regarding compute and resources used, indicating that the hardware specifications are not detailed in the provided main text.
Software Dependencies No No specific software dependencies with version numbers are listed in the provided text.
Experiment Setup Yes Our MDP setup exactly replicates that of [24], with N = 5000, λ = 1, µ = 1; see the appendix for further details. We run all estimators over 100 separate trajectories of length t = 104N of the above MDP initialized in its stationary distribution. Each estimator was run over 50 independent simulator trajectories, each over 3 105 requests. The DQ and OPE estimators shared a common linear approximation architecture with basis functions that count the number of drivers at every occupancy level.