Diogo R. Ferreira

From Chessprogramming wiki
Jump to: navigation, search

Home * People * Diogo R. Ferreira

Diogo R. Ferreira [1]

Diogo R. Ferreira,
a Portuguese computer scientist and professor of Information systems at Instituto Superior Técnico, Technical University of Lisbon. His research interests include data science, deep learning, GPU computing, data mining and process mining applied to nuclear fusion, plasma tomography, and disruption prediction [2].

Chess Ratings

In 2010, Diogo R. Ferreira participated on a chess ratings competition offered by Kaggle [3] and made it to 4th place [4] . His approach is based on the Bradley-Terry model, but with a slightly different interpretation of what the strength of a player is, and with a custom procedure for estimating that strength [5] . In his ICGA Journal paper, Determining the Strength of Chess Players based on actual Play [6] , Ferreira elaborates on strength estimation based on a per move basis with the help of a strong chess program and distribution of gain. A gain was defined by the score difference of Houdini's 1.5 fixed depth 20 ply search of all consecutive positions along the game-record, alternating assigned to the player who made the move. Four algorithms for different purposes were defined:

  1. Determine the relative strength between two players i and j.
  2. Determine the relative strength of each player i, in terms of a perceived Elo rating r'i for a set of players in a tournament, where the actual Elo ratings ri are known.
  3. Determine the engine strength based on the rating differences between the engine and a set of players whose Elo ratings and/or perceived ratings are known.
  4. Determine the strength of a player after a number of moves, based on the rating difference between player and engine, when an estimate of engine strength is available.

Ratings of various chess players and programs were estimated based on the distribution of gains from various games, The Game of the Century between Donald Byrne and the 13 year old Bobby Fischer was given as sample for algorithm 1 with game score differences, yielding in a rating difference of 113 points in favor to Fischer. The London Chess Classic 2011 [7] [8] was used as sample for algorithm 2 and 3, while algorithm 4 was applied for Kasparov versus Deep Blue 1997, which, despite Garry Kasparov lost the match, did result in a higher rating for him - 2754 versus 2741 for Deep Blue.

Selected Publications

[9] [10] [11]

2000 ...

2010 ...

External Links

References

Up one level