Struct minimax::strategies::mcts::MonteCarloTreeSearch

pub struct MonteCarloTreeSearch<G: Game> { /* private fields */ }

A strategy that uses random playouts to explore the game tree to decide on the best move. This can be used without an Evaluator, just using the rules of the game.

Implementations

impl<G: Game> MonteCarloTreeSearch<G>

pub fn new(options: MCTSOptions) -> Self

pub fn new_with_policy( options: MCTSOptions, policy: Box<dyn RolloutPolicy<G = G> + Sync> ) -> Self

Create a searcher with a custom rollout policy. You could bias the random move generation to prefer certain kinds of moves, always choose winning moves, etc.

pub fn set_max_rollouts(&mut self, rollouts: u32)

Instead of a timeout, run this many rollouts to choose a move.

Trait Implementations

impl<G> Strategy<G> for MonteCarloTreeSearch<G>where G: Sync + Game, G::S: Clone + Send, G::M: Copy + Sync,

fn choose_move(&mut self, s: &G::S) -> Option<G::M>

fn set_timeout(&mut self, timeout: Duration)

For strategies that can ponder indefinitely, set the timeout. This can be changed between calls to choose_move.

fn set_max_depth(&mut self, depth: u8)

Set the maximum depth to evaluate (instead of the timeout). This can be changed between calls to choose_move.

fn principal_variation(&self) -> Vec<G::M>

From the last choose_move call, return the principal variation, i.e. the best sequence of moves for both players.