Struct A2cConfig 

pub struct A2cConfig {
    pub learning_rate: f64,
    pub gamma: f64,
    pub gae_lambda: f64,
    pub value_coef: f64,
    pub entropy_coef: f64,
    pub n_steps: usize,
    pub num_envs: usize,
    pub max_grad_norm: f64,
    pub normalize_advantages: bool,
    pub seed: u64,
}

A2C configuration parameters.

These hyperparameters control the synchronous Advantage Actor-Critic training process. Default values target classic discrete-control tasks like CartPole: a single gradient update per n_steps-long rollout collected across num_envs synchronous actors, full n-step returns (gae_lambda = 1.0), and global gradient-norm clipping.

Fields

learning_rate: f64

Learning rate for the shared actor-critic optimizer.

gamma: f64

Discount factor (gamma).

gae_lambda: f64

GAE lambda parameter. The classic A2C default of 1.0 recovers full n-step returns (no GAE smoothing).

value_coef: f64

Value function loss coefficient.

entropy_coef: f64

Entropy bonus coefficient.

n_steps: usize

Rollout length: number of environment steps collected per update (Mnih et al. 2016 use 5-step rollouts).

num_envs: usize

Number of parallel synchronous actors.

max_grad_norm: f64

Maximum global gradient norm for clipping.

normalize_advantages: bool

Whether to normalize advantages to zero-mean/unit-variance per rollout before computing the policy loss.

seed: u64

Seed threaded into seeded network init (e.g. MlpBurnConfig::with_seed) for reproducibility. Two trainers built with the same seed produce bit-identical initialization.

Implementations

impl A2cConfig

pub fn new() -> Self

Create a new default configuration.

pub fn validate(&self) -> Result<()>

Validate configuration parameters.

Returns an Err describing the first invalid field encountered.

pub fn learning_rate(self, lr: f64) -> Self

Set the learning rate.

pub fn gamma(self, gamma: f64) -> Self

Set the discount factor gamma.

pub fn gae_lambda(self, lambda: f64) -> Self

Set the GAE lambda parameter.

pub fn value_coef(self, coef: f64) -> Self

Set the value function loss coefficient.

pub fn entropy_coef(self, coef: f64) -> Self

Set the entropy bonus coefficient.

pub fn n_steps(self, steps: usize) -> Self

Set the rollout length (steps collected per update).

pub fn num_envs(self, envs: usize) -> Self

Set the number of parallel synchronous actors.

pub fn max_grad_norm(self, norm: f64) -> Self

Set the maximum global gradient norm for clipping.

pub fn normalize_advantages(self, enabled: bool) -> Self

Enable or disable per-rollout advantage normalization.

pub fn seed(self, seed: u64) -> Self

Set the reproducibility seed.

Trait Implementations

impl Clone for A2cConfig

fn clone(&self) -> A2cConfig

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for A2cConfig

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

Formats the value using the given formatter.

impl Default for A2cConfig

fn default() -> Self

Returns the “default value” for a type.