Trait AgentSimulator 

pub trait AgentSimulator: Send + Sync {
    // Required methods
    fn get_num_actions(&self) -> usize;
    fn get_num_observation_bits(&self) -> usize;
    fn get_num_reward_bits(&self) -> usize;
    fn horizon(&self) -> usize;
    fn max_reward(&self) -> Reward;
    fn min_reward(&self) -> Reward;
    fn model_update_action(&mut self, action: Action);
    fn gen_percept_and_update(&mut self, bits: usize) -> u64;
    fn model_revert(&mut self, steps: usize);
    fn gen_range(&mut self, end: usize) -> usize;
    fn gen_f64(&mut self) -> f64;
    fn boxed_clone_with_seed(&self, seed: u64) -> Box<dyn AgentSimulator>;

    // Provided methods
    fn observation_stream_len(&self) -> usize { ... }
    fn observation_key_mode(&self) -> ObservationKeyMode { ... }
    fn observation_repr_from_stream(
        &self,
        observations: &[PerceptVal],
    ) -> Vec<PerceptVal>  { ... }
    fn reward_offset(&self) -> i64 { ... }
    fn get_explore_exploit_ratio(&self) -> f64 { ... }
    fn discount_gamma(&self) -> f64 { ... }
    fn boxed_clone(&self) -> Box<dyn AgentSimulator> { ... }
    fn norm_reward(&self, reward: f64) -> f64 { ... }
    fn gen_percepts_and_update(&mut self) -> (Vec<PerceptVal>, Reward) { ... }
}

Interface for an agent that can be simulated during MCTS.

This trait allows the MCTS algorithm to interact with an agent (like Agent in agent.rs) to perform “imagined” actions and receive “imagined” percepts during planning.

Required Methods

fn get_num_actions(&self) -> usize

Returns the number of possible actions the agent can perform.

fn get_num_observation_bits(&self) -> usize

Returns the bit-width used to encode observations.

fn get_num_reward_bits(&self) -> usize

Returns the bit-width used to encode rewards.

fn horizon(&self) -> usize

Returns the planning horizon (depth of simulations).

fn max_reward(&self) -> Reward

Returns the maximum possible reward value.

fn min_reward(&self) -> Reward

Returns the minimum possible reward value.

fn model_update_action(&mut self, action: Action)

Updates the internal model state with a simulated action.

fn gen_percept_and_update(&mut self, bits: usize) -> u64

Generates a simulated percept and updates the model state.

fn model_revert(&mut self, steps: usize)

Reverts the model state to a previous point in the simulation.

fn gen_range(&mut self, end: usize) -> usize

Generates a random value in [0, end).

fn gen_f64(&mut self) -> f64

Generates a random f64 in [0, 1).

fn boxed_clone_with_seed(&self, seed: u64) -> Box<dyn AgentSimulator>

Creates a boxed clone of this simulator, re-seeding any RNG state.

Provided Methods

fn observation_stream_len(&self) -> usize

Returns the number of observation symbols per action.

fn observation_key_mode(&self) -> ObservationKeyMode

Returns the observation key mode for search-tree branching.

fn observation_repr_from_stream( &self, observations: &[PerceptVal], ) -> Vec<PerceptVal>

Returns the observation representation used for tree branching.

fn reward_offset(&self) -> i64

Returns the reward offset used to ensure encoded rewards are non-negative.

Paper-compatible encoding uses unsigned reward bits and shifts rewards by an offset.

fn get_explore_exploit_ratio(&self) -> f64

Returns the exploration-exploitation constant (often denoted as C).

fn discount_gamma(&self) -> f64

Returns the discount factor for future rewards.

fn boxed_clone(&self) -> Box<dyn AgentSimulator>

Creates a boxed clone of this simulator for parallel search.

fn norm_reward(&self, reward: f64) -> f64

Normalizes a reward value to [0, 1] based on the agent’s range and horizon.

For discounted rewards, the cumulative range is sum_{t=0}^{h-1} gamma^t * (max - min). Similarly, the minimum cumulative reward is sum_{t=0}^{h-1} gamma^t * min.

fn gen_percepts_and_update(&mut self) -> (Vec<PerceptVal>, Reward)

Helper to generate a percept stream, update the model, and return a search key + reward.

Implementors

impl AgentSimulator for Agent