Struct oxymcts::DefaultLazyTreePolicy

pub struct DefaultLazyTreePolicy<State: GameTrait, EV: Evaluator<State, Reward, A>, A: Clone + Default, Reward: Clone> { /* fields omitted */ }

Explores at least once each child node, before going deeper.

Implementations

impl<State: GameTrait, EV: Evaluator<State, Reward, A, Args = f64>, A: Clone + Default, Reward: Clone> DefaultLazyTreePolicy<State, EV, A, Reward> where
    Reward: Div + ToPrimitive + Add + Zero, 

pub fn select(
    tree: &mut LazyMctsTree<State, Reward, A>,
    turn: &State::Player,
    evaluator_args: &EV::Args
) -> NodeId

pub fn expand(
    node_to_expand: NodeMut<'_, LazyMctsNode<State, Reward, A>>,
    root_state: State
) -> (NodeId, State)

Trait Implementations

impl<State, EV, A, Reward> LazyTreePolicy<State, EV, A, Reward> for DefaultLazyTreePolicy<State, EV, A, Reward> where
    State: GameTrait,
    Reward: Clone + Div + Add + ToPrimitive + Zero,
    EV: Evaluator<State, Reward, A, Args = f64>,
    A: Clone + Default, 

fn tree_policy(
    tree: &mut LazyMctsTree<State, Reward, A>,
    root_state: State,
    evaluator_args: &EV::Args
) -> (NodeId, State)

Choose the best node, for example we apply the UCT to choose the best node then we expand it and we return the expansion.

fn update_state(root_state: State, historic: &[State::Move]) -> State

This method is only needed because we don't store the state in each node so we need, to update the state with the stored historic in each node before simulating or expanding it.

fn best_child(
    tree: &LazyMctsTree<State, Reward, A>,
    turn: &State::Player,
    parent_id: NodeId,
    eval_args: &EV::Args
) -> NodeId

This method use the Evaluator to get best child using evaluate_child.