Pattern Classification in No-Limit Poker: A Head-Start Evolutionary Approach

Abstract

Abstract. We have constructed a poker classification system which makes informed betting decisions based upon three defining features extracted while playing poker: hand value, risk, and aggressiveness. The system is implemented as a poker player agent, and as such, the goals of the classifier are not only to correctly determine whether each hand should be folded, called, or raised, but to win as many chips as possible from the other players. The decision space is found by evolutionary methods, starting from a designed initial state. Our results showed that evolving an agent from a data-driven "head-start" position resulted in the best performance over agents evolved from scratch, random agents, data-driven agents, and "always fold" agents (a surprisingly effective strategy).

Building a No Limit Texas Hold’em Poker Agent Based on Game Logs Using Supervised Learning

“…More recently, it should be emphasized the article [9], which describes a new and more effective technique for building counter strategies based on Restricted Nash Response. Some achievements were also made using machine learning classifiers like in [10] where were studied evolutionary methods to classify in game actions.…”

Algorithms for Evolving No-Limit Texas Hold'em Poker Playing Agents

“…No-Limit Texas Hold'em Poker agents were developed in (Beattie et al, 2007), and were capable of playing large-scale games with up to ten players at a table, and tournaments with hundreds of tables. Evolutionary methods were used to evolve twodimensional matrices corresponding to the current game state.…”

“…That is, all players can see all information relevant to the game state at all times. Recently, games of imperfect information, such as Poker (Barone and While, 1999;Beattie et al, 2007;Billings et al, 2002;Johanson, 2007) has started to attract attention in the research community. Unlike Chess and Checkers, there are certain, where all information is available to all players, Poker involves deception and hidden information.…”

“…This variant limits the size of the decision space by limiting the potential decisions available to an agent. Another variant, known as No-Limit Texas Hold'em (Beattie et al, 2007;Booker, 2004), changes only one rule, but results in many more potential decisions for an agent, and consequently, a much larger decision space.…”

“…Rather than reduce the decision space, we use evolutionary algorithms (Samuel, 1959;Schaeffer et al, 1992;Thrun, 1995;Pollack and Blair, 1998;Barone and While, 1999;Kendall and Whitwell, 2001;Lubberts and Miikkulainen, 2001;Tesauro, 2002;Hauptman and Sipper, 2005;Beattie et al, 2007) to teach our agents a guided path to a good solution. Evolutionary algorithms mimic natural evolution, and reward good decisions while punishing less desirable ones.…”