Spaces

class jaxatari.spaces.Box(low: float | ~numpy.ndarray | ~jax.Array, high: float | ~numpy.ndarray | ~jax.Array, shape: ~typing.Tuple[int, ...] | None = None, dtype: ~numpy.dtype = <class 'jax.numpy.float32'>)

Bases: Space

A jittable n-dimensional box space.

This space represents the Cartesian product of n closed intervals. Each interval has its own lower and upper bound.

It can be initialized in two ways: 1. With scalar bounds and an explicit shape, creating a box with uniform bounds.

Example: Box(low=0.0, high=1.0, shape=(3, 4))

2. With array-like bounds, where the shape is inferred from the bounds arrays. Example: Box(low=jnp.array([0., -1.]), high=jnp.array([1., 1.]))

contains(x: Array) → Array

Check if a point x is contained within the box.

range() → tuple[Array, Array]

Returns the lower and upper bounds of the space.

sample(key: Array) → Array

Generates a random sample from the space.

The sample is uniformly distributed over the box.

class jaxatari.spaces.Dict(spaces: dict)

Bases: Space

A jittable dictionary of simpler jittable spaces (Pytree container).

contains(x: dict) → Array

Check whether the given Pytree is contained in the space.

sample(key: Array) → OrderedDict

class jaxatari.spaces.Discrete(num_categories: int)

Bases: Space

Minimal jittable class for discrete spaces.

contains(x: Array) → Array

Check whether specific object is within space.

range() → tuple[float, float]

sample(key: Array) → Array

Sample random action uniformly from set of categorical choices.

class jaxatari.spaces.Space

Bases: object

Minimal jittable class for abstract spaces.

contains(x: Array) → Any

range()

sample(key: Array) → Array

class jaxatari.spaces.Tuple(spaces: Sequence[Space])

Bases: Space

A jittable tuple of simpler jittable spaces (Pytree container).

contains(x: tuple) → Array

Check whether the given Pytree is contained in the space.

sample(key: Array) → tuple

Sample a random tuple from all subspaces.

jaxatari.spaces.stack_space(space: Space, stack_size: int) → Space

Recursively wraps a space or a Pytree of spaces to add a stacking dimension to each leaf space. Handles Box and Discrete spaces as leaves.