Ordinal Potential-based Player Rating (2306.05366v4)
Abstract: It was recently observed that Elo ratings fail at preserving transitive relations among strategies and therefore cannot correctly extract the transitive component of a game. We provide a characterization of transitive games as a weak variant of ordinal potential games and show that Elo ratings actually do preserve transitivity when computed in the right space, using suitable invertible mappings. Leveraging this insight, we introduce a new game decomposition of an arbitrary game into transitive and cyclic components that is learnt using a neural network-based architecture and that prioritises capturing the sign pattern of the game, namely transitive and cyclic relations among strategies. We link our approach to the known concept of sign-rank, and evaluate our methodology using both toy examples and empirical data from real-world games.
- Sign rank versus VC dimension. In Proc. of COLT, volume 49, pages 47–80.
- Open-ended learning in symmetric zero-sum games. In Proc. of ICML, volume 97, pages 434–443.
- Re-evaluating evaluation. In Proc. of NeurIPS, pages 3272–3283.
- On the limitations of Elo: Real-world games, are transitive, not additive. In Proc. of AISTATS.
- Borkar, V. S. (1997). Stochastic approximation with two time scales. Systems & Control Letters.
- Flows and Decompositions of Games: Harmonic and Potential Games. Mathematics of Operations Research, 36(3):474–503.
- Real world games look like spinning tops. In Proc. of NeurIPS.
- Elo, A. (1961). The USCF rating system. Chess Life.
- Elo, A. (1978). The rating of Chess players, past and present. Ishi Press International.
- Greub, W. H. (1975). Linear algebra. Springer Science & Business Media, 4thsuperscript4𝑡ℎ4^{th}4 start_POSTSUPERSCRIPT italic_t italic_h end_POSTSUPERSCRIPT Edition.
- Statistical ranking and combinatorial hodge theory. Math. Program., 127(1):203–244.
- Nfgtransformer: Equivariant representation learning for normal-form games. In ICLR.
- Potential Games. Games and Economic Behavior, 14:124–143.
- The sign-rank of AC0. SIAM Journal on Computing, 39(5):1833–1855.
- Sismanis, Y. (2010). How I won the “Chess ratings - Elo vs the rest of the world” Competition. CoRR, abs/1012.4571.
- The network HHD: quantifying cyclic competition in trait-performance models of tournaments. SIAM Rev., 64(2):360–391.