readme

Matlab/Octave package: BNPPL
Bayesian nonparametric Plackett-Luce for ranking data

Download software

Link to the demo (html)

Description

This Matlab/Octave package implements sampling algorithms for simulation and Bayesian inference in (mixture of) nonparametric Plackett-Luce models (Caron et al., 2014). These models are useful for analysing partial ranking data consisting of ordered lists of top-m items among a very large, potentially unbounded set of items.

The package has been tested on Matlab R2014a (with statistics toolbox) and Octave 3.6.4.

F. Caron, Y.W. Teh, T.B. Murphy. Bayesian nonparametric Plackett-Luce models for the analysis of preferences for college degree programmes. The Annals of Applied Statistics, vol. 8, no2, pp. 1145-1181, 2014. Download paper.

Contents

Demo of the package

demo_nppl.m: reproduces some figures of the paper and provides a demo of the functions in the package - see the htlm version of the script

Sampler and inference in nonparametric Plackett-Luce

sample_nppl.m: samples from the nonparametric Plackett-Luce model
mcmc_nppl.m: runs a MCMC algorithm for the nonparametric Plackett-Luce model

Sampler and inference in mixture of nonparametric Plackett-Luce

sample_mixnppl.m: samples from the nonparametric mixture of Plackett-Luce model
mcmc_mixnppl.m: runs a MCMC algorithm for the nonparametric mixture of Plackett-Luce model

Processing of the outputs of the MCMC algorithms

coclustering.m: computes the coclustering/similarity matrix from the output of the MCMC algorithm
clust_est_binder.m: gets a Bayesian point estimate of the partition from the MCMC output using Binder loss function

Latest corrections

Last update: 04-07-2014. Correction of a bug in clust_est_binder.m

Permission is granted for anyone to copy, use, or modify these programs and accompanying documents for purposes of research or education, 
provided this copyright notice is retained, and note is made of any changes that have been made.

These programs and documents are distributed without any warranty, express or implied.  
As the programs were written for research purposes only, they have not been tested to the degree that would be advisable in any important application.  
All use of these programs is entirely at the user's own risk.