7

After reading this paper about Monte Carlo methods for imperfect information games with elements of uncertainty, I couldn't understand the application of the determinization step in the author's implementation of the algorithm for the Knockout game.

Determinization is defined as the transformation from an instance of an imperfect information game to ab instance of a perfect one. It means that all players should see the cards of each other after the determinization step.

Why can't the players see the cards of each other in the code above?

nbro
  • 39,006
  • 12
  • 98
  • 176
tamirok
  • 71
  • 1

2 Answers2

1

MCTS only need to "see" states in respect of reward. All game mechanics is abstarcted away from MCTS and MCTS only access actions and rewards. MCTS player don't access states itself, it's only choose action according to backpropagated reward. For partially observed MCTS player can't even access rewards of states, but instead access only expected reward over information set. Because player don't see reward of each state of information set but only expected reward over all set he can't knowledgeably choose specific state from information set. Player choose random state from information set according to some distribution instead. That mean player "don't see" which state from information is set actually realized.

mirror2image
  • 695
  • 5
  • 14
1

All the methods in the GameState class that is used to represent state, are stubs, and without these, the MCTS algorithm won't do anything at all. In particular, the DoMove method just changes who's turn it is, without actually taking any action.

Probably the reason the players can't see each other's cards is that this is not a completed implementation. Someone is either still working on this, or gave up half way.

John Doucette
  • 9,147
  • 1
  • 17
  • 52