Reinforcement learning in Tarock

Požrl, Domen (avtor), Sadikov, Aleksander (mentor)

In this thesis we study and analyse the effectiveness of Q-learning when it comes to learning and playing the card game Tarock. We present many problems that are inherently present in every attempt at mastering a card game with artificial intelligence. We developed several models, each with a distinct way of looking at the game and distinct behaviour in different game scenarios. We designed different testing conditions ranging from playing against theoretical players and other Tarock playing software, to playing against real people. We evaluated our models under real life and computational conditions. Our best performing model was able to compete with, and on many occasions beat an average Tarock player.

• Jezik: Angleški jezik
• Ključne besede: reinforcement learning, Tarock, cards, game playing
• Vrsta gradiva: Magistrsko delo/naloga
• Tipologija: 2.09 - Magistrsko delo
• Organizacija: FRI - Fakulteta za računalništvo in informatiko
• Leto izida: 2021
• PID: 20.500.12556/RUL-133070
• COBISS.SI-ID: 87302915
• Datum objave v RUL: 10.11.2021

Sekundarni jezik

• Jezik: Slovenski jezik
• Naslov: Spodbujevalno učenje v taroku

Izvleček

V tem magistrskem delu preučimo in analiziramo učinkovitost metode Q-learning v kontekstu učenja in igranja igre tarok. Predstavimo probleme, ki se pojavljajo pri vsakem poskusu igranja iger s kartami s pomočjo umetne inteligence. Razvili smo več modelov, vsakega z drugačnim pogledom na igro in drugačnim odgovorom na različna stanja igre. Zasnovali smo več testnih okolij, od igranja proti teoretičnim igralcem in drugim programom za igranje taroka, do igranja proti pravim ljudem. Modele smo ovrednotili pod življenjskimi in računskimi pogoji. Naš najboljši model se je uspešno boril in večkrat celo premagoval povprečne igralce taroka.

• Ključne besede: spodbujevalno učenje, tarok, karte, igranje iger