Struct NeuralRLAgent 

pub struct NeuralRLAgent {
    pub config: AdvancedConfig,
    pub learning_rate: f64,
    pub strategy: ExplorationStrategy,
    /* private fields */
}

Adaptive (bandit-style) reinforcement-learning agent for algorithm/strategy selection.

Despite the historical “Neural” name (kept for API stability), this is a classical multi-armed-bandit reinforcement learner – epsilon-greedy, UCB, an approximate Thompson-sampling-style rule, or adaptive-uncertainty exploration, per the configured ExplorationStrategy – not a deep neural network. It genuinely tracks per-arm reward statistics ((pull_count, mean_reward)) and adapts its choices using real randomness (scirs2_core::random::rng); nothing here is a fixed placeholder. See AdvancedProcessor::execute_adaptive for the intended usage (choosing among several candidate implementations of the same operation).

Fields

config: AdvancedConfig

Agent configuration

learning_rate: f64

Learning rate: how quickly record_reward updates track new observations. 0.0 means “plain running average of every reward ever observed”; a positive rate makes it an exponential moving average that favors recent observations.

strategy: ExplorationStrategy

Exploration strategy used by Self::select_arm

Implementations

impl NeuralRLAgent

pub fn new(config: AdvancedConfig, strategy: ExplorationStrategy) -> Self

Creates a new agent using the given exploration strategy; config.learning_rate seeds Self::learning_rate.

pub fn arm_count(&self) -> usize

Number of arms this agent currently has statistics for.

pub fn arm_stats(&self, arm: usize) -> Option<(u64, f64)>

The current (pull_count, mean_reward) estimate for arm, or None if it has never been selected.

pub fn select_arm(&mut self, n_arms: usize) -> usize

Selects one of n_arms candidate actions according to the configured ExplorationStrategy and this agent’s accumulated reward history. Any arm never pulled before is always tried first (standard bandit warm-up), before the configured strategy takes over.

pub fn record_reward(&mut self, arm: usize, reward: f64)

Records an observed reward for arm, updating its running statistics (growing the tracked arm count if needed).

Trait Implementations

impl Clone for NeuralRLAgent

fn clone(&self) -> NeuralRLAgent

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for NeuralRLAgent

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for NeuralRLAgent

fn default() -> Self

Returns the “default value” for a type.