Reinforcement Learning agents playing Hearthstone: The influence of state description when learning in a large, partially observable state space

Tutea, Ionuț-Alexandru (2020) Reinforcement Learning agents playing Hearthstone: The influence of state description when learning in a large, partially observable state space. Bachelor's Thesis, Artificial Intelligence.

Preview	Text AI_BA_2020_IONUTALEXANDRUTUTEA.pdf Download (479kB) | Preview
	Text toestemming.pdf Restricted to Registered users only Download (100kB)

Abstract

Many real world problems cannot be easily formalized into compact states that fully describe the real state of the world. It is thus important to look at how different ways of representing a state would influence the performance of AI agents. In this research we compare the performance of reinforcement learning agents which learn using different game state representations when playing "Hearthstone", an online collectible card game which has a state space size much larger than complex games such as Chess. Moreover, its states are only partially observable by the player and randomness governs over certain events. It is found that, when playing with very simple decks that do not require complex strategies, agents which have a simpler state description have a win rate of about 81% against a random player, whereas agents which have a larger state description only win 78% of the time regardless of the learning algorithm used. However, when playing with more complex decks that require synergistic actions in order to maximize their performance, agents with a simpler state description performed worse than before while the agents with larger state descriptions performed better with win rates of 88% against a random agent.

Item Type:	Thesis (Bachelor's Thesis)
Supervisor name:	Wiering, M.A.
Degree programme:	Artificial Intelligence
Thesis type:	Bachelor's Thesis
Language:	English
Date Deposited:	10 Feb 2020 12:42
Last Modified:	10 Feb 2020 12:42
URI:	https://fse.studenttheses.ub.rug.nl/id/eprint/21526

Actions (login required)

View Item