Struct Demonstrations 

pub struct Demonstrations { /* private fields */ }

A fixed dataset of expert (observation, action) pairs for supervised imitation learning.

Observations are stored as a single flat row-major Vec<f32> of shape [len, obs_dim]; the discrete expert action for example i lives at actions[i]. The dataset is immutable once constructed and serves host-side minibatches by index via Demonstrations::batch.

Implementations

impl Demonstrations

pub fn new(obs: Vec<f32>, actions: Vec<i64>, obs_dim: usize) -> Result<Self>

Build a dataset from a flat observation buffer and a parallel action vector.

obs must be row-major of length len * obs_dim and actions must have length len, where len is inferred from actions. Returns an Err if the lengths are inconsistent or obs_dim is zero.

Arguments

• obs - Flat row-major observations of shape [len, obs_dim].
• actions - Expert action index per example.
• obs_dim - Dimensionality of a single observation.

pub fn len(&self) -> usize

Number of (observation, action) examples in the dataset.

pub fn is_empty(&self) -> bool

Whether the dataset contains no examples.

pub fn obs_dim(&self) -> usize

Dimensionality of a single observation.

pub fn batch<B: Backend>( &self, indices: &[usize], device: &B::Device, ) -> (Tensor<B, 2>, Tensor<B, 1, Int>)

Gather the given example indices into a minibatch of tensors.

Returns (obs, actions) where obs has shape [k, obs_dim] and actions has shape [k], with k = indices.len(). Gathering is done host-side; the resulting tensors are plain inputs/labels with no gradient tracking concerns.

Panics

Panics if any index is out of bounds (>= len).

Arguments

• indices - Example indices to gather into the minibatch.
• device - Backend device the tensors are allocated on.

Trait Implementations

impl Clone for Demonstrations

fn clone(&self) -> Demonstrations

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for Demonstrations

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

Formats the value using the given formatter.