A Profitable Online Poker Agent

• Autor(a) principal: João Pedro Almeida Campos
• Data de Publicação: 2013
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: https://hdl.handle.net/10216/112267
• topic: Engenharia electrotécnica, electrónica e informática; Electrical engineering, Electronic engineering, Information engineering

Games of incomplete information, such as poker, are a continuous source of research and study in the area of artificial intelligence. Poker presents challenging problems such as opponent modeling, risk management and bluff detection. The development of agents capable of probabilistic calculations considering those problems is considered to be difficult to achieve, since dynamic adaption is required in order to create a robust computer poker player. This thesis focuses on the development of a poker agent able to play against human players and aiming to achieve the dynamic adaptation needed to beat some human players online. This will be achieved by using some sets of information about each player the agent plays against. Using Holdem Manager, a tool that registers the hands played in an online poker room; it is possible to obtain statistics about every player the agent is playing against. The agent is able to explore some of these statistics so that it can better decide on which action to take. Some factors like how aggressive an opponent is, the position held at the table, how many players are involved, how much money is involved, and the hand dealt to the agent are a few portions of the information sets used to compute the agent's behavior. This agent was developed based on a short-stack strategy, and through the use of the sets of information provided by the Holdem Manager. For the first time in the Computer Poker literature, results on online Poker agent games versus human players in a controlled environment are presented, and without the players being aware their opponent was a computer agent. The agent is able to play live online poker versus human players, and presents a small profit in the No-Limit Texas Hold'em poker game at micro stakes, namely 0.02 and 0.01 cents.