|
Poker MCTS
A Pokerbotting library
|
Monte Carlo Tree Search or short MCTS is a simulation based search algorithm that can be used to solve extensive-form games with imperfect information. Complex games like NLTH have a large number of states due to the element of chance and the number of opponents. Instead of searching the complete state space (which is intractable) MCTS only samples one possible outcome per iteration. As the number of iterations increases, the true value of a state can be approximated. MCTS is a anytime algorithm which means that the computation can be interrupted at any time to retrieve a valid result (or a approximation of it if not enough iterations have been performed.) The search is performed on a tree containing the following types of nodes:
Root node is the start of the tree. The node represents the state we want to evaluate.
Action nodes are nodes in which the same player has to make a decision as in the root node.
Opponent nodes represent the opposing players decisions.
Terminal nodes represent either nodes in which the game ended by folding by all but one player or by going to a showdown.
Chance nodes are used in places where chance makes a move (draw a card).
Each iteration begins at the root node and recursively selects children according to a selection strategy that chooses which action to take. If the state for the selected action does not exist, the last state is expanded and it’s children are added to the tree before continuing the selection until a terminal history is reached. This history is then evaluated according to a simulation strategy that calculates the expected value. In the final step of the iteration the calculated value is propagated back up the tree by an backpropagation strategy. Figure 3.13 shows a graphical representation of one iteration. After a certain amount of time or number of predefined iterations a final move strategy selects the actual actions to play from the root node.
1.8.15