Guess It is a simple card game of bluffing and opponent modelling designed by Rufus Isaacs of the Rand Corporation. In this paper, we discuss the technical details needed to equip an adaptive learning algorithm with the ability to play the game and report a series of experiments that compare the performance of different learning techniques. Our results show that in most cases the different techniques produce perfect countering strategies against a number of fixed opponents, although there are differences in the speed of learning and robustness to change between the different algorithms. We further report experiments where the learning techniques compete against each other in a coadaptive setting.
展开▼