RewardTransform

Trait RewardTransform 

Source
pub trait RewardTransform: Send + Sync {
    // Required methods
    fn transform(
        &self,
        rewards: &[f64],
        context: &RewardContext<'_>,
    ) -> Result<Vec<f64>, RloxError>;
    fn name(&self) -> &str;
}
Expand description

Trait for composable reward transformations.

Each transform takes raw rewards and a context and produces shaped rewards. Transforms can be chained: ClipReward -> PBRSTransform -> ScaleReward.

Required Methods§

Source

fn transform( &self, rewards: &[f64], context: &RewardContext<'_>, ) -> Result<Vec<f64>, RloxError>

Transform a batch of rewards.

Source

fn name(&self) -> &str

Human-readable name.

Implementors§