Caron 2012 - Bayesian Inference for Generalized Bradley-Terry Models

Routing Summary

Full capture of Caron & Doucet (JCGS 2012; INRIA RR-7445 / arXiv:1011.1761): EM and Gamma-Gibbs algorithms for Bradley-Terry, home advantage, ties, team comparisons and Plackett-Luce, plus identifiability, hyperparameters and experiments. 6 notes.

Concept Map

ConceptNoteTypeDepends OnKey Result
BT log-likelihood, latent Gamma Caron 2012 - Bradley-Terry Likelihood and Latent Exponential AugmentationtheoremOverviewMM = EM; Gamma Gibbs
Home advantage Caron 2012 - Home Advantage, Ties and Team ComparisonstheoremBT Gamma conditional
Rao-Kupper tiesCaron 2012 - Home Advantage, Ties and Team ComparisonstheoremBT conditional is a Gamma mixture
Team BTCaron 2012 - Home Advantage, Ties and Team ComparisonstheoremBTwinner-id latent
Plackett-LuceCaron 2012 - Plackett-Luce Model for Full RankingstheoremBTstage-wise exponentials
Dirichlet-Gamma identifiabilityCaron 2012 - Identifiability, Hyperparameters and Further ExtensionstheoremBTonly identified; rescale step
NASCAR / chessCaron 2012 - Experiments NASCAR and ChessexampleallGibbs beats M-H; Bayes beats MAP

Notes

Sources

Caron and Doucet 2012 - Efficient Bayesian Inference for Generalized Bradley-Terry Models

See Also

Discrete Choice Models · Reward Modeling from Human Preferences · Hierarchical Models