Struct arbor::MCTS

pub struct MCTS<P: Player, A: Action, S: GameState<P, A>> {
    pub info: Info,
    /* private fields */
}

This struct is the main launch point for this crate. It holds the state of execution for the MCTS algorithm. Use it’s associated methods to operate the search and tune performance.

Fields

info: Info

Provides metrics about the shape and size of the game tree. For informational purposes only.

Implementations

impl<P: Player, A: Action, S: GameState<P, A>> MCTS<P, A, S>

pub fn new(root: S) -> Self

Call this method to instantiate a new search with default parameters. The root game state from which to search is passed as a value to be owned by the MCTS struct.

pub fn best(&self) -> Option<A>

Pick the best move after some time spent pondering. Returns None if ponder has not yet been called.

pub fn ply<F>(&self, f: &mut F) where
    F: FnMut((A, f32, f32)), 

Iterate through the actions in the first ply. The callback f is called for each action in the first ply with a tuple of (a, w, s) where a is the action, w is the expected value of the action, and s is the confidence in the value of the action. s is similar to standard deviation where closer to zero is more confident.

pub fn ponder(&mut self, n: usize)

Call this method to search the root game state a given number of iterations. This method may be called any number of times to improve the search results. Call ply or best to get the current search results.

impl<P: Player, A: Action, S: GameState<P, A>> MCTS<P, A, S>

pub fn with_exploration(self, exploration: f32) -> Self

Sets the exploration parameter. This sets the balance between exploration and exploitation when MCTS determines which action to choose. Set to a value > 0.

pub fn with_expansion_minimum(self, expansion: u32) -> Self

Sets the expansion parameter. This is the minimum number of times a leaf node should be visited before expanding it into a branch node.

pub fn with_custom_evaluation(self) -> Self

Enables the custom evaluation method.

pub fn with_transposition(self) -> Self

Enables transposition detection. Experimental.

pub fn with_entropy(self) -> Self

Seeds the internal random number generator from entropy. This is inteneded to produce non-deterministic search results.