For a good introduction to the most basic variant, Glicko-1, see this document.īoth the Elo and Glicko systems model players' skill through a "rating" score in such a way that if two players P1 and P2 have ratings $r_1$ and $r_2$ prior to playing each other, then the expected probability that P1 defeats P2 is a logistic function of the ratings difference $r_1 - r_2$. The International Chess Federation (FIDE) uses the Elo system to rate its players, while many chess websites (including ) use variants of the more advanced Glicko rating system. The primary systems used to rate players and predict the outcomes of their games are Elo and Glicko. Our production model makes significant improvements upon widely-used existing models see the "Results" section below for details. We task ourselves with predicting the outcome of a new player's 2nd-ever game on the online chess platform Lichess using only data from that player's 1st game along with metadata about the upcoming game and about the new player's opponent's last game. We use the chess engine Stockfish to evaluate players' moves and positions, then we feed these evaluations, along with other metadata about the players and their games, to a neural network model.Ĭonsidering the 2-week timeframe of this project and the heavy computational resources required, our production model is a proof-of-concept aimed at solving a simple instance of the general problem of better predictions of game outcomes for new players. In addressing this problem, our primary innovation - as compared to existing player rating / matchmaking systems like Elo and Glicko - is to look at not only the outcomes of players' games but also the moves that the players made during their games. To achieve better matchmaking, models need to be able to make good probabilistic predictions of the outcomes of future games: if they can make good predictions, then a good match can be made by pairing two players such that the predicted probability of either one winning is close to 50%. In this project, we create models aimed at better matchmaking for new players on online chess platforms.
0 Comments
Leave a Reply. |
Details
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |