Analyzing Simulations in Monte-Carlo Tree Search for the Game of Go

In Monte-Carlo Tree Search, simulations play a crucial role since they replace the evaluation function used in classical game-tree search and guide the development of the game tree. Despite their importance, not too much is known about the details of how they work. This paper starts a more in-depth study of simulations, using the game of Go, and in particular the program Fuego, as an example. Playout policies are investigated in terms of the number of blunders they make, and in terms of how many points they lose over the course of a simulation. The result is a deeper understanding of the different components of the Fuego playout policy, as well as an analysis of the shortcomings of current methods for evaluating playouts.