Struct CandleDQN 

pub struct CandleDQN { /* private fields */ }

Implementations

impl CandleDQN

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

Trait Implementations

impl QFunction for CandleDQN

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

Compute Q-values for all actions. Returns [batch_size, n_actions].

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

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

fn td_step( &mut self, obs: &TensorData, actions: &TensorData, targets: &TensorData, weights: Option<&TensorData>, ) -> Result<(f64, TensorData), NNError>

Perform one DQN TD gradient step. Returns (loss, td_errors) where td_errors can be used for PER.

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

Compute target Q-values using the target network.

fn hard_update_target(&mut self)

Hard-copy parameters to the target network.

impl Send for CandleDQN

impl Sync for CandleDQN