Struct ReplayBatch 

pub struct ReplayBatch {
    pub observations: Vec<f32>,
    pub actions: Vec<i64>,
    pub rewards: Vec<f32>,
    pub next_observations: Vec<f32>,
    pub dones: Vec<bool>,
    pub obs_dim: usize,
}

One minibatch sampled from the replay buffer.

All fields are CPU-side primitive vectors; convert to Burn tensors via ReplayBatch::to_burn_tensors when handing them to the trainer.

Fields

observations: Vec<f32>

Flattened current observations, shape [batch_size * obs_dim].

actions: Vec<i64>

Actions taken, length batch_size.

rewards: Vec<f32>

Rewards received, length batch_size.

next_observations: Vec<f32>

Flattened next observations, shape [batch_size * obs_dim].

dones: Vec<bool>

Episode-end mask, length batch_size. true means the transition terminated the episode (so the TD target drops the bootstrap term).

obs_dim: usize

Length of one observation slice (so to_tensors can reshape).

Implementations

impl ReplayBatch

pub fn len(&self) -> usize

Number of transitions in the batch.

pub fn is_empty(&self) -> bool

true if the batch is empty.

pub fn to_burn_tensors<B: Backend>( &self, device: &B::Device, ) -> ReplayBurnTensors<B>

Stack the batch into Burn tensors on device.

Returns a named ReplayBurnTensors struct so trainers can pattern-match named fields rather than positional tuple elements.

Shapes (all on device):

• observations: [batch, obs_dim], f32
• actions: [batch], i64 (Burn Int kind)
• rewards: [batch], f32
• next_observations: [batch, obs_dim], f32
• dones: [batch], f32 (0.0 / 1.0; the TD-target formula (1 - done) consumes it directly as a float mask).

Trait Implementations

impl Clone for ReplayBatch

fn clone(&self) -> ReplayBatch

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for ReplayBatch

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

Formats the value using the given formatter.