Struct CandleTwinQ 

pub struct CandleTwinQ { /* private fields */ }

Implementations

impl CandleTwinQ

pub fn new( obs_dim: usize, act_dim: usize, hidden: usize, lr: f64, device: Device, ) -> Result<Self, NNError>

pub fn q1_forward(&self, obs: &Tensor, actions: &Tensor) -> Result<Tensor>

Forward through Q1 with autograd preserved (for actor loss in SAC/TD3).

pub fn twin_q_forward( &self, obs: &Tensor, actions: &Tensor, ) -> Result<(Tensor, Tensor)>

Forward through both Q-networks with autograd preserved (for SAC actor loss).

Trait Implementations

impl ContinuousQFunction for CandleTwinQ

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

Compute Q-value for (obs, action). Returns [batch_size].

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

Compute twin Q-values for (obs, action). Returns (q1, q2), each [batch_size].

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

Compute target twin Q-values (from target networks).

fn critic_step( &mut self, obs: &TensorData, actions: &TensorData, targets: &TensorData, ) -> Result<TrainMetrics, NNError>

Perform one TD gradient step on both critics.

fn soft_update_targets(&mut self, tau: f32)

Polyak soft update of target networks.

impl Send for CandleTwinQ

impl Sync for CandleTwinQ