Effective Sampling and Learning for Mallows Models with Pairwise-Preference Data
Authors: Tyler Lu, Craig Boutilier
JMLR 2014 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments on real-world data sets demonstrate the effectiveness of our approach. |
| Researcher Affiliation | Academia | Tyler Lu EMAIL Craig Boutilier EMAIL Department of Computer Science University of Toronto 6 King s College Rd. Toronto, ON, Canada M5S 3G4 |
| Pseudocode | Yes | Algorithm 1 AMP Approximate Mallows Posterior Algorithm 2 MMP Sample Mallows Posterior using Metropolis Algorithm 3 SP: Sample Mallows Mixture Posterior using Gibbs Algorithm 4 Local Kemeny |
| Open Source Code | No | The paper does not explicitly state that the source code for the methodology is openly available or provide a link to a repository. It mentions a C++ implementation in a footnote but no access information. |
| Open Datasets | Yes | We apply our EM algorithm to a subset of the Movielens data set (see www.grouplens.org) The Sushi data set consists of 5000 complete rankings over 10 varieties of sushi indicating sushi preferences (Kamishima et al., 2005). |
| Dataset Splits | Yes | The Sushi data set... We used 3500 preferences for training and 1500 for validation. Movielens data set... use 3986 preferences for training and 1994 for validation. To test posterior prediction performance, we use 1000 complete rankings, distinct from both the training and validation sets |
| Hardware Specification | Yes | The C++ implementation of our algorithms have EM wall clock times of 15–20 minutes (Intel Xeon dual-core, 3GHz) |
| Software Dependencies | No | The paper mentions a "C++ implementation" but does not specify any particular software libraries or their version numbers. |
| Experiment Setup | Yes | In each experiment, we generate random model parameters as follows: π is drawn from a Dirichlet distribution with a uniform parameter vector of 5s; σ is drawn uniformly at random; and φ values are drawn uniformly at random from [0.2, 0.8]. Log-likelihoods are approximated using our Monte Carlo estimator (with K T = 120). |