| Abstract: |  Research concerning games with imperfect information has received a lot of attention lately. The reasons behind this phenomenon are wide applications of its results in the real world and its high di culty. There exists a lot of algorithms solving games with imperfect-formation.
Problem is that each of these algorithms was tested on a dierent game and therefore their comparison is complicated. The main goal of this thesis is to implement three algorithms, namely Counterfactual Regret Minimization, Information Set Search and Monte-Carlo Tree Search, and compare their results on three dierent games with imperfect information. The games chosen for this task are Latent Tic-tac-toe, Leduc holdem poker and a version of pursuit-evasion game Border Patrol. Each of them contains a dierent problem for algorithms to deal with. Results of this bachelor thesis shown that Counterfactual Regret Minimization outperforms rest of the algorithms in all games and that results of Information Set Search
are strongly dependent on the model chosen in a speci c game.
|
|---|