Casper : Design and Development of a Case - Based Poker Player

Poker provides a challenging domain for Artificial Intelligence research due to the game’s properties such as hidden information (the other player’s cards) and nondeterminism (random shuffling of the deck). Recent approaches to Poker research have required intensive knowledge engineering efforts. This thesis discusses the design and development of a CASe-based Poker playER (CASPER) that uses the Case-Based Reasoning methodology to make betting decisions at the poker table. The results suggest it is possible to record instances of games played between strong poker players and then reuse these to obtain a similar performance therefore bypassing the need for the initial, intensive knowledge engineering process. An investigation into deriving optimal feature weights using evolutionary algorithms has also been conducted. Casper has been extensively evaluated by challenging various sets of opponents, including both computerised opponents and real opponents.