Optimal, approximately optimal, and fair play of the fowl play card game

Abstract

After introducing the jeopardy card game Fowl Play, we present equations for optimal two-player play, describe their solution with a variant of value iteration, and visualize the optimal play policy. Next, we discuss the approximation of optimal play and note that neural network learning can achieve a win rate within 1 % of optimal play yet with a 5-orders-of-magnitude reduction in memory requirements. Optimal komi (i.e., compensation points) are computed for the two-player games of Pig and Fowl Play. Finally, we make use of such komi computations in order to redesign Fowl Play for two-player fairness, creating the game Red Light. © 2014 Springer International Publishing Switzerland.