Struct MonteCarloTreeSearch 

pub struct MonteCarloTreeSearch<T: Board, K: RandomGenerator> { /* private fields */ }

The main struct for running the Monte Carlo Tree Search algorithm.

It holds the search tree, the random number generator, and the configuration for the search.

Implementations

impl<T: Board, K: RandomGenerator> MonteCarloTreeSearch<T, K>

pub fn builder(board: T) -> MonteCarloTreeSearchBuilder<T, K>

Returns a new builder for MonteCarloTreeSearch.

Examples found in repository

examples/tic_tac_toe.rs (line 12)

fn main() {
    // Create a new Tic-Tac-Toe board
    let board = TicTacToeBoard::default();

    // Create a new MCTS search instance
    let mut mcts = MonteCarloTreeSearch::builder(board)
        .with_alpha_beta_pruning(false)
        .with_random_generator(CustomNumberGenerator::default())
        .build();

    // Run the search for 20,000 iterations
    mcts.iterate_n_times(20000);

    // Print the chances
    let root = mcts.get_root();
    for node in root.children() {
        println!(
            "Move: {:?} = {:.2?}%",
            node.value().prev_move,
            node.value().wins_rate() * 100.0
        );
    }

    // Get the most promising move
    let best_move_node = root.get_best_child().unwrap();
    let best_move = best_move_node.value().prev_move;

    println!("The best move is: {:?}", best_move);
    assert_eq!(best_move, Some(4));
}

pub fn new(board: T, rg: K, use_alpha_beta_pruning: bool) -> Self

Creates a new MonteCarloTreeSearch instance.

It is recommended to use the builder pattern via MonteCarloTreeSearch::builder() instead.

pub fn get_tree(&self) -> &Tree<MctsNode<T>>

Returns an immutable reference to the underlying search tree.

pub fn get_next_mcts_action(&self) -> &MctsAction

Returns the next MCTS action to be performed. Useful for debugging and visualization.

pub fn execute_action(&mut self)

Executes a single step of the MCTS algorithm (Selection, Expansion, Simulation, or Backpropagation).

pub fn do_iteration(&mut self) -> Vec<NodeId>

Performs one full iteration of the MCTS algorithm (Selection, Expansion, Simulation, Backpropagation). Returns the path of nodes that were updated during backpropagation.

pub fn iterate_n_times(&mut self, n: u32)

Runs the MCTS search for a specified number of iterations.

Examples found in repository

examples/tic_tac_toe.rs (line 18)

fn main() {
    // Create a new Tic-Tac-Toe board
    let board = TicTacToeBoard::default();

    // Create a new MCTS search instance
    let mut mcts = MonteCarloTreeSearch::builder(board)
        .with_alpha_beta_pruning(false)
        .with_random_generator(CustomNumberGenerator::default())
        .build();

    // Run the search for 20,000 iterations
    mcts.iterate_n_times(20000);

    // Print the chances
    let root = mcts.get_root();
    for node in root.children() {
        println!(
            "Move: {:?} = {:.2?}%",
            node.value().prev_move,
            node.value().wins_rate() * 100.0
        );
    }

    // Get the most promising move
    let best_move_node = root.get_best_child().unwrap();
    let best_move = best_move_node.value().prev_move;

    println!("The best move is: {:?}", best_move);
    assert_eq!(best_move, Some(4));
}

pub fn get_root(&self) -> MctsTreeNode<'_, T>

Returns a reference to the root node of the search tree.

Examples found in repository

examples/tic_tac_toe.rs (line 21)

fn main() {
    // Create a new Tic-Tac-Toe board
    let board = TicTacToeBoard::default();

    // Create a new MCTS search instance
    let mut mcts = MonteCarloTreeSearch::builder(board)
        .with_alpha_beta_pruning(false)
        .with_random_generator(CustomNumberGenerator::default())
        .build();

    // Run the search for 20,000 iterations
    mcts.iterate_n_times(20000);

    // Print the chances
    let root = mcts.get_root();
    for node in root.children() {
        println!(
            "Move: {:?} = {:.2?}%",
            node.value().prev_move,
            node.value().wins_rate() * 100.0
        );
    }

    // Get the most promising move
    let best_move_node = root.get_best_child().unwrap();
    let best_move = best_move_node.value().prev_move;

    println!("The best move is: {:?}", best_move);
    assert_eq!(best_move, Some(4));
}

impl<T: Board> MonteCarloTreeSearch<T, StandardRandomGenerator>

pub fn from_board(board: T) -> Self