Computing Parametric Ranking Models via Rank-Breaking

Authors: Hossein Azari Soufiani, David Parkes, Lirong Xia

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results are presented to show the computational efficiency along with statistical performance of the proposed method. We conduct experimental studies to compare our algorithm to the MC-EM algorithm for RUMs. We consider RUMs with normal distributions and study running time and Kendall correlation. Experimental results show that our algorithm runs much faster than the MC-EM algorithm while achieving comparable, and sometimes even better Kendall correlation.
Researcher Affiliation Academia Hossein Azari Soufiani AZARI@FAS.HARVARD.EDU David C. Parkes PARKES@EECS.HARVARD.EDU Harvard University, 33 Oxford Street, Cambridge, MA 02138 USA Lirong Xia XIAL@CS.RPI.EDU Rensselaer Polytechnic Institute, Troy, NY 12180, USA
Pseudocode Yes Algorithm 1 GMMG(Dr) For all a, a , compute Xa a G (Dr). Compute GMMG(Dr) according to (2) using the moment conditions in (3) (e.g. using gradient descent). return GMMG(Dr).
Open Source Code Yes The code is provided in the R package Stat Rank (Chen & Azari Soufiani, 2013). We plan to extend the algorithms and analysis to partial orders, non-location families such as RUMs parameterized by mean and variance, and to GRUMs (Azari Soufiani et al., 2013c) and GRUMs with multiple types (Azari Soufiani et al., 2013b). URL http://cran.r-project. org/web/packages/Stat Rank/index.html.
Open Datasets No The synthetic datasets are generated as follows. Let m = 5. The ground truth γ is generated from the Dirichlet distribution Dirichlet( 1) which is a distribution on an m dimensional unit simplex. Then, for any given γ we generate up to n = 200 full rankings from the location family with normal distributions. This describes how data was generated, but does not provide access information to a public dataset.
Dataset Splits No The paper mentions generating up to n = 200 full rankings but does not specify any training, validation, or test dataset splits or cross-validation setup.
Hardware Specification Yes All experiments are run on a 2.4 Ghz, Intel Core 2 duo 32 bit laptop.
Software Dependencies No The code is provided in the R package Stat Rank (Chen & Azari Soufiani, 2013). While an R package is mentioned, specific version numbers for R or any dependent libraries are not provided.
Experiment Setup No The paper describes how synthetic datasets were generated (e.g., m=5, n=200 rankings), but it does not specify concrete experimental setup details such as hyperparameters (learning rates, batch sizes, epochs) for the algorithms used (GMM or MC-EM).