Trait DeterministicPolicy 

pub trait DeterministicPolicy {
    // Required methods
    fn act(&self, obs: &TensorData) -> Result<TensorData, NNError>;
    fn target_act(&self, obs: &TensorData) -> Result<TensorData, NNError>;
    fn soft_update_target(&mut self, tau: f32);
    fn learning_rate(&self) -> f32;
    fn set_learning_rate(&mut self, lr: f32);
    fn save(&self, path: &Path) -> Result<(), NNError>;
    fn load(&mut self, path: &Path) -> Result<(), NNError>;
}

Deterministic policy for TD3.

Training steps (td3_actor_step) are intentionally NOT on this trait because they require autograd to flow through the critic’s Q-network. Trait methods convert tensors to TensorData (Vec), severing the computation graph. Use the backend-specific inherent td3_actor_step method instead.

Required Methods

fn act(&self, obs: &TensorData) -> Result<TensorData, NNError>

Compute deterministic action. Returns [batch_size, act_dim].

fn target_act(&self, obs: &TensorData) -> Result<TensorData, NNError>

Compute target policy action (from target network).

fn soft_update_target(&mut self, tau: f32)

Polyak soft update of target network.

fn learning_rate(&self) -> f32

fn set_learning_rate(&mut self, lr: f32)

fn save(&self, path: &Path) -> Result<(), NNError>

fn load(&mut self, path: &Path) -> Result<(), NNError>

Implementors