Scenario#

protocol enki_env.Scenario#

A scenario is a generator of world.

It exposes a callable that accepts a random seed and returns a pyenki.World.

>>> world = scenario(seed=0)

The callable has an optional argument to copy the random generator from another world, which is passed during the environment reset with seed=None.

Users have two ways to implement a scenario:

The can write a callable that conforms to Scenario.__call__(), where they must pass the random seed to the world and possibly copy the random generator, e.g.:

def my_scenario(seed: int, copy_rng_from: pyenki.World = None) -> pyenki.World:
    world = pyenki.World(seed=seed)
    if copy_rng_from:
        world.copy_random_generator(copy_rng_from)
    ...
    return world

They can specialize a sub-class of BaseScenario, implementing one or both of BaseScenario.make() and BaseScenario.init(). This class takes care of setting random seeds and generators in its implementation of BaseScenario.__call__().

For example

class MyScenario

    def make(self) -> pyenki.World:
        return pyenki.World(radius=100)

    def init(self, pyenki.World: world) -> None:
        robot = pyenki.Thymio2()
        robot.angle = world.random_generator.uniform(0, math.pi * 2)
        world.add_object(world)

Users should use pyenki.World.random_generator for random sampling to ensure reproducibility.

Classes that implement this protocol must have the following methods / attributes:

__call__(seed: int, /, copy_rng_from: pyenki.World | None = None) → pyenki.World#

Generates of a world with a given random seed.

Seed:: The random seed.
Copy_rng_from:: Another world to inherit the random generator from, if provided.
Returns:: The generated world.

class enki_env.BaseScenario#

Base class for implementations of enki_env.Scenario protocol.

Users can specialize sub-classes by implementing one or both of BaseScenario.make() and BaseScenario.init().

For example

class MyScenario

    def make(self) -> pyenki.World:
        return pyenki.World(radius=100)

    def init(self, pyenki.World: world) -> None:
        robot = pyenki.Thymio2()
        robot.angle = world.random_generator.uniform(0, math.pi * 2)
        world.add_object(world)

__call__(seed: int, /, copy_rng_from: pyenki.World | None = None) → pyenki.World#

Implements Scenario protocol.

calls BaseScenario.make()
sets the random seed and generator
calls BaseScenario.init()

Seed:: The random seed.
Copy_rng_from:: Another world to inherit the random generator from, if provided.
Returns:: The world.

init(world: pyenki.World) → None#

Virtual method that initializes a given world.

The base implementation does nothing. Can be overridden by sub-classes to add objects.

Important

Any random sampling should use pyenki.World.random_generator to ensure reproducibility.

Parameters:: world – The generated world

make() → pyenki.World#

Virtual method that returns a world.

The base implementation returns a empty world with no boundaries. Can be overridden by sub-classes to add a boundary.

Important

Should defer random sampling to BaseScenario.init().

Seed:: the random seed.
Returns:: the world.

Scenario

Contents

Scenario#