Module connect6::policy

Predefined policy for agent to play game.

Policy represents algorithm that make choice in given situation. This module provides 5 predefined policies, AlphaZero policy, random policy, io policy, multi-policy and pure MCTS policy.

• Policy : trait for playing game with Agent.
• AlphaZero : implementation of policy AlphaZero.
• RandomPolicy : select possible position with randomness.
• IoPolicy : read user input.
• MultiPolicy : Black-White seperable policy, pass two different policies as initialize parameter.
• DefaultPolicy : Pure Monte Carlo tree search implementation.

Examples

struct ZeroPolicy {}
impl Policy for ZeroPolicy {
    fn next(&mut self, game: &Game) -> Option<(usize, usize)> {
        Some((0, 0))
    }
}

let mut policy = ZeroPolicy {};
let result = Agent::new(&mut policy).play();
assert!(result.is_err());

Structs

AlphaZero	Implementation of policy AlphaZero based on combined MCTS with non-linear value approximator.
DefaultPolicy	Policy for pure Monte Carlo tree search implementation
HyperParameter	Hyperparameter for implementing AlphaZero.
IoPolicy	Policy for in-game user selection with io
MultiPolicy	Black-White seperable policy
Node	Shared memory for making simulation wihout copying board and possible selections.
RandomEvaluator	Evaluator for test, Random Value Evaluator
RandomPolicy	Play game with random selection.
Simulate	Game simulator with shared memory for efficient tree searching

Traits

Evaluator	Evaluator for applying value, policy approximator to AlphaZero.
Policy	trait for playing game with Agent.

Functions

diff_board	compare the board and return the difference between position by position.
hash	generate hash value of board