Reinforcement Learning in Card Games

Fokus ovoga rada je istražiti primjenjivost algoritama podržanog učenja na kartaškim igrama. Rad sadrži kratki uvod u podržano učenje i pojašnjava najčešće korištene algoritme u podržanom učenju. Na primjeru kartaške igre "Bela" je izgrađen simulacijski okvir s mogućnostima testiranja raznih agenata. U sklopu simulacije je implementiran agent koji koristi q učenje za razvijanje politike. Navedeni agent je testiran u velikom broju epizoda protiv ostalih agenata nakon čega je provedena analiza uspješnosti agenta.The focus of this paper is to investigate the applicability of reinforcement learning algorithms on card games. The paper contains a brief introduction to reinforcement learning and explains the most common algorithms used in it. A simulation framework for the card game "Bela" was built with the possibility of testing various agents. A machine learning agent, that uses q learning for policy building was also developed. The said agent was tested in a large number of episodes against other agents after which its performance was analyzed in detail