Ranking from pairwise comparisons

Demo

• use https://www.opinionx.co/ to choose the most interesting LD topic

Applications

• ranking anime https://manitary.github.io/MAL-ranking/

Given comparisons determine rank

• Elo, Glicko
  • https://ratings.fide.com/
  • https://www.goratings.org/en/
• TrueSkill, approximates the posterior using Expectation Propagation
• A Bayesian Approximation Method for Online Ranking
• SVM Ranking
  • c_ij is an observed comparison, 1 or 0
  • x_ij = e_j - e_i (e_i is 1 at the ith coordinate)
  • y_ij = 2c_ij - 1 (-1 or 1)
• https://github.com/erensezener/rank-centrality
• http://choix.lum.li/en/latest/index.html
• https://github.com/Janzert/halite_ranking
• https://github.com/OpenDebates/openskill.py

Time varying

• https://www.remi-coulom.fr/WHR/
• https://github.com/lucasmaystre/kickscore

Active ranking: determine what comparisons to perform

• https://proceedings.mlr.press/v70/maystre17a.html proves that Quicksort works well under the Bradley-Terry model
• https://github.com/gfxdisp/asap based on expected information gain
  • uses TrueSkill, should be faster with Weng-Lin
• https://github.com/ipsl/Ada-IIR

Related

• Dueling bandits