Monte-Carlo Tree Search for the Game of "7 Wonders"

Abstract : Monte-Carlo Tree Search algorithm, and in particular with the Upper Confidence Bounds formula, provided huge improvements for AI in numerous games, particularly in Go, Hex, Havannah, Amazons and Breakthrough. In this work we study this algorithm on a more complex game, the game of " 7 Wonders ". This card game gathers together several known difficult features, such as hidden information, N-player and stochasticity. It also includes an inter-player trading system that induces a combinatorial search to decide which decisions are legal. Moreover it is difficult to hand-craft an efficient evaluation function since the card values are heavily dependent upon the stage of the game and upon the other player decisions. We show that, in spite of the fact that " 7 Wonders " is apparently not so related to abstract games, a lot of known results still hold.
Type de document :
Communication dans un congrès
European Conference in Artificial Intelligence (ECAI), Aug 2014, Prague, Czech Republic. pp.64 - 77, 2014, Computer Game Workshop 〈10.1007/978-3-319-14923-3_5〉
Liste complète des métadonnées

Littérature citée [23 références]  Voir  Masquer  Télécharger

https://hal.inria.fr/hal-01406496
Contributeur : Fabien Teytaud <>
Soumis le : jeudi 1 décembre 2016 - 11:35:25
Dernière modification le : vendredi 2 décembre 2016 - 01:00:30
Document(s) archivé(s) le : mardi 21 mars 2017 - 06:39:35

Fichier

7wonders.pdf
Fichiers produits par l'(les) auteur(s)

Identifiants

Collections

Citation

Denis Robilliard, Cyril Fonlupt, Fabien Teytaud. Monte-Carlo Tree Search for the Game of "7 Wonders". European Conference in Artificial Intelligence (ECAI), Aug 2014, Prague, Czech Republic. pp.64 - 77, 2014, Computer Game Workshop 〈10.1007/978-3-319-14923-3_5〉. 〈hal-01406496〉

Partager

Métriques

Consultations de la notice

67

Téléchargements de fichiers

242