TY - GEN
T1 - Designing Card Game Strategies with Genetic Programming and Monte-Carlo Tree Search
T2 - 2020 IEEE Symposium Series on Computational Intelligence, SSCI 2020
AU - Chia, Hao Cheng
AU - Yeh, Tsung Su
AU - Chiang, Tsung Che
N1 - Publisher Copyright:
© 2020 IEEE.
PY - 2020/12/1
Y1 - 2020/12/1
N2 - This paper addresses an agent design problem of a digital collectible card game, Hearthstone, which is a two-player turn-based game. The agent has to play cards based on the game state, the hand cards, and the deck of cards to defeat the opponent. First, we design a rule-based agent by searching for the board evaluation criterion through genetic programming (GP). Then, we integrate the rule-based agent into the Monte-Carlo tree search (MCTS) framework to generate an advanced agent. Performance of the proposed agents are verified by playing against three participants in two recent Hearthstone competitions. Experimental results showed that the GP-agent can beat a simple MCTS agent and the mid-level agent in the competition. The MCTS-GP agent showed competitive performance against the best agents in the competition. We also examine the rule found by GP and observed that GP is able to identify key attributes of game states and to combine them into a useful rule automatically.
AB - This paper addresses an agent design problem of a digital collectible card game, Hearthstone, which is a two-player turn-based game. The agent has to play cards based on the game state, the hand cards, and the deck of cards to defeat the opponent. First, we design a rule-based agent by searching for the board evaluation criterion through genetic programming (GP). Then, we integrate the rule-based agent into the Monte-Carlo tree search (MCTS) framework to generate an advanced agent. Performance of the proposed agents are verified by playing against three participants in two recent Hearthstone competitions. Experimental results showed that the GP-agent can beat a simple MCTS agent and the mid-level agent in the competition. The MCTS-GP agent showed competitive performance against the best agents in the competition. We also examine the rule found by GP and observed that GP is able to identify key attributes of game states and to combine them into a useful rule automatically.
KW - Hearthstone: Heroes of Warcraft
KW - Monte-Carlo tree search
KW - collectible card games
KW - genetic programming
UR - http://www.scopus.com/inward/record.url?scp=85099684464&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85099684464&partnerID=8YFLogxK
U2 - 10.1109/SSCI47803.2020.9308459
DO - 10.1109/SSCI47803.2020.9308459
M3 - Conference contribution
AN - SCOPUS:85099684464
T3 - 2020 IEEE Symposium Series on Computational Intelligence, SSCI 2020
SP - 2351
EP - 2358
BT - 2020 IEEE Symposium Series on Computational Intelligence, SSCI 2020
PB - Institute of Electrical and Electronics Engineers Inc.
Y2 - 1 December 2020 through 4 December 2020
ER -