Identifying Strategies in <i>Dominion</i> Using Playtrace Clustering

We demonstrate the use of playtraces and playtrace clustering to identify strategies and card synergies in deck building card games, using <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Dominion</i> as an example. We analyze playtraces generated from both online human play and a variety of AI agents, examining two types: card counts by round in a player's deck and N-Grams generated from player actions. Using both the <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$L_{k}$</tex-math> </inline-formula>-norm and Jensen–Shannon distance measures, in-conjunction with <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$K$</tex-math> </inline-formula>-Means, <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$K$</tex-math> </inline-formula>-Medoids and DBSCAN algorithms, we show that playtraces and distinct clusters can reveal both longer term strategies and card synergies. In addition, we use a restricted play framework to increase the variation in strategies and tactics explored by AI agents. Finally, we suggest that the game-agnostic, N-Gram-based approach may support strategy exploration in tabletop games more broadly.