Files
Gymnasium/gym/core.py

345 lines
11 KiB
Python
Raw Normal View History

from __future__ import annotations
from abc import abstractmethod
from typing import TypeVar, Generic, Tuple, Union, Optional, SupportsFloat
2018-02-26 17:35:07 +01:00
import gym
from gym import spaces
from gym.utils import seeding
Seeding update (#2422) * Ditch most of the seeding.py and replace np_random with the numpy default_rng. Let's see if tests pass * Updated a bunch of RNG calls from the RandomState API to Generator API * black; didn't expect that, did ya? * Undo a typo * blaaack * More typo fixes * Fixed setting/getting state in multidiscrete spaces * Fix typo, fix a test to work with the new sampling * Correctly (?) pass the randomly generated seed if np_random is called with None as seed * Convert the Discrete sample to a python int (as opposed to np.int64) * Remove some redundant imports * First version of the compatibility layer for old-style RNG. Mainly to trigger tests. * Removed redundant f-strings * Style fixes, removing unused imports * Try to make tests pass by removing atari from the dockerfile * Try to make tests pass by removing atari from the setup * Try to make tests pass by removing atari from the setup * Try to make tests pass by removing atari from the setup * First attempt at deprecating `env.seed` and supporting `env.reset(seed=seed)` instead. Tests should hopefully pass but throw up a million warnings. * black; didn't expect that, didya? * Rename the reset parameter in VecEnvs back to `seed` * Updated tests to use the new seeding method * Removed a bunch of old `seed` calls. Fixed a bug in AsyncVectorEnv * Stop Discrete envs from doing part of the setup (and using the randomness) in init (as opposed to reset) * Add explicit seed to wrappers reset * Remove an accidental return * Re-add some legacy functions with a warning. * Use deprecation instead of regular warnings for the newly deprecated methods/functions
2021-12-08 22:14:15 +01:00
from gym.logger import deprecation
from gym.utils.seeding import RandomNumberGenerator
ObsType = TypeVar("ObsType")
ActType = TypeVar("ActType")
2016-04-27 08:00:58 -07:00
class Env(Generic[ObsType, ActType]):
"""The main OpenAI Gym class. It encapsulates an environment with
2016-04-28 10:33:37 -07:00
arbitrary behind-the-scenes dynamics. An environment can be
partially or fully observed.
The main API methods that users of this class need to know are:
step
[WIP] add support for seeding environments (#135) * Make environments seedable * Fix monitor bugs - Set monitor_id before setting the infix. This was a bug that would yield incorrect results with multiple monitors. - Remove extra pid from stats recorder filename. This should be purely cosmetic. * Start uploading seeds in episode_batch * Fix _bigint_from_bytes for python3 * Set seed explicitly in random_agent * Pass through seed argument * Also pass through random state to spaces * Pass random state into the observation/action spaces * Make all _seed methods return the list of used seeds * Switch over to np.random where possible * Start hashing seeds, and also seed doom engine * Fixup seeding determinism in many cases * Seed before loading the ROM * Make seeding more Python3 friendly * Make the MuJoCo skipping a bit more forgiving * Remove debugging PDB calls * Make setInt argument into raw bytes * Validate and upload seeds * Skip box2d * Make seeds smaller, and change representation of seeds in upload * Handle long seeds * Fix RandomAgent example to be deterministic * Handle integer types correctly in Python2 and Python3 * Try caching pip * Try adding swap * Add df and free calls * Bump swap * Bump swap size * Try setting overcommit * Try other sysctls * Try fixing overcommit * Try just setting overcommit_memory=1 * Add explanatory comment * Add what's new section to readme * BUG: Mark ElevatorAction-ram-v0 as non-deterministic for now * Document seed * Move nondetermistic check into spec
2016-05-29 09:07:09 -07:00
reset
2016-04-28 10:33:37 -07:00
render
close
[WIP] add support for seeding environments (#135) * Make environments seedable * Fix monitor bugs - Set monitor_id before setting the infix. This was a bug that would yield incorrect results with multiple monitors. - Remove extra pid from stats recorder filename. This should be purely cosmetic. * Start uploading seeds in episode_batch * Fix _bigint_from_bytes for python3 * Set seed explicitly in random_agent * Pass through seed argument * Also pass through random state to spaces * Pass random state into the observation/action spaces * Make all _seed methods return the list of used seeds * Switch over to np.random where possible * Start hashing seeds, and also seed doom engine * Fixup seeding determinism in many cases * Seed before loading the ROM * Make seeding more Python3 friendly * Make the MuJoCo skipping a bit more forgiving * Remove debugging PDB calls * Make setInt argument into raw bytes * Validate and upload seeds * Skip box2d * Make seeds smaller, and change representation of seeds in upload * Handle long seeds * Fix RandomAgent example to be deterministic * Handle integer types correctly in Python2 and Python3 * Try caching pip * Try adding swap * Add df and free calls * Bump swap * Bump swap size * Try setting overcommit * Try other sysctls * Try fixing overcommit * Try just setting overcommit_memory=1 * Add explanatory comment * Add what's new section to readme * BUG: Mark ElevatorAction-ram-v0 as non-deterministic for now * Document seed * Move nondetermistic check into spec
2016-05-29 09:07:09 -07:00
seed
2016-04-27 08:00:58 -07:00
And set the following attributes:
action_space: The Space object corresponding to valid actions
observation_space: The Space object corresponding to valid observations
reward_range: A tuple corresponding to the min and max possible rewards
2016-04-27 08:00:58 -07:00
Note: a default reward range set to [-inf,+inf] already exists. Set it if you want a narrower range.
The methods are accessed publicly as "step", "reset", etc...
2016-04-27 08:00:58 -07:00
"""
2021-07-29 02:26:34 +02:00
2016-04-27 08:00:58 -07:00
# Set this in SOME subclasses
2021-07-29 02:26:34 +02:00
metadata = {"render.modes": []}
reward_range = (-float("inf"), float("inf"))
spec = None
2016-04-27 08:00:58 -07:00
# Set these in ALL subclasses
action_space: spaces.Space[ActType]
observation_space: spaces.Space[ObsType]
2016-04-27 08:00:58 -07:00
Seeding update (#2422) * Ditch most of the seeding.py and replace np_random with the numpy default_rng. Let's see if tests pass * Updated a bunch of RNG calls from the RandomState API to Generator API * black; didn't expect that, did ya? * Undo a typo * blaaack * More typo fixes * Fixed setting/getting state in multidiscrete spaces * Fix typo, fix a test to work with the new sampling * Correctly (?) pass the randomly generated seed if np_random is called with None as seed * Convert the Discrete sample to a python int (as opposed to np.int64) * Remove some redundant imports * First version of the compatibility layer for old-style RNG. Mainly to trigger tests. * Removed redundant f-strings * Style fixes, removing unused imports * Try to make tests pass by removing atari from the dockerfile * Try to make tests pass by removing atari from the setup * Try to make tests pass by removing atari from the setup * Try to make tests pass by removing atari from the setup * First attempt at deprecating `env.seed` and supporting `env.reset(seed=seed)` instead. Tests should hopefully pass but throw up a million warnings. * black; didn't expect that, didya? * Rename the reset parameter in VecEnvs back to `seed` * Updated tests to use the new seeding method * Removed a bunch of old `seed` calls. Fixed a bug in AsyncVectorEnv * Stop Discrete envs from doing part of the setup (and using the randomness) in init (as opposed to reset) * Add explicit seed to wrappers reset * Remove an accidental return * Re-add some legacy functions with a warning. * Use deprecation instead of regular warnings for the newly deprecated methods/functions
2021-12-08 22:14:15 +01:00
# Created
_np_random: RandomNumberGenerator | None = None
@property
def np_random(self) -> RandomNumberGenerator:
"""Initializes the np_random field if not done already."""
if self._np_random is None:
self._np_random, seed = seeding.np_random()
return self._np_random
Seeding update (#2422) * Ditch most of the seeding.py and replace np_random with the numpy default_rng. Let's see if tests pass * Updated a bunch of RNG calls from the RandomState API to Generator API * black; didn't expect that, did ya? * Undo a typo * blaaack * More typo fixes * Fixed setting/getting state in multidiscrete spaces * Fix typo, fix a test to work with the new sampling * Correctly (?) pass the randomly generated seed if np_random is called with None as seed * Convert the Discrete sample to a python int (as opposed to np.int64) * Remove some redundant imports * First version of the compatibility layer for old-style RNG. Mainly to trigger tests. * Removed redundant f-strings * Style fixes, removing unused imports * Try to make tests pass by removing atari from the dockerfile * Try to make tests pass by removing atari from the setup * Try to make tests pass by removing atari from the setup * Try to make tests pass by removing atari from the setup * First attempt at deprecating `env.seed` and supporting `env.reset(seed=seed)` instead. Tests should hopefully pass but throw up a million warnings. * black; didn't expect that, didya? * Rename the reset parameter in VecEnvs back to `seed` * Updated tests to use the new seeding method * Removed a bunch of old `seed` calls. Fixed a bug in AsyncVectorEnv * Stop Discrete envs from doing part of the setup (and using the randomness) in init (as opposed to reset) * Add explicit seed to wrappers reset * Remove an accidental return * Re-add some legacy functions with a warning. * Use deprecation instead of regular warnings for the newly deprecated methods/functions
2021-12-08 22:14:15 +01:00
@abstractmethod
def step(self, action: ActType) -> Tuple[ObsType, float, bool, dict]:
2016-04-28 10:33:37 -07:00
"""Run one timestep of the environment's dynamics. When end of
episode is reached, you are responsible for calling `reset()`
to reset this environment's state.
2016-04-27 08:00:58 -07:00
Accepts an action and returns a tuple (observation, reward, done, info).
2016-04-27 08:00:58 -07:00
Args:
2019-05-24 14:29:01 -07:00
action (object): an action provided by the agent
2016-04-27 08:00:58 -07:00
Returns:
observation (object): agent's observation of the current environment
reward (float) : amount of reward returned after previous action
done (bool): whether the episode has ended, in which case further step() calls will return undefined results
2022-02-13 01:39:03 +01:00
info (dict): contains auxiliary diagnostic information (helpful for debugging, logging, and sometimes learning)
2016-04-27 08:00:58 -07:00
"""
raise NotImplementedError
2016-04-27 08:00:58 -07:00
@abstractmethod
def reset(
self,
*,
seed: Optional[int] = None,
return_info: bool = False,
options: Optional[dict] = None,
) -> Union[ObsType, tuple[ObsType, dict]]:
"""Resets the environment to an initial state and returns an initial
observation.
2022-02-13 01:39:03 +01:00
This method should also reset the environment's random number
generator(s) if `seed` is an integer or if the environment has not
yet initialized a random number generator. If the environment already
has a random number generator and `reset` is called with `seed=None`,
the RNG should not be reset.
Moreover, `reset` should (in the typical use case) be called with an
integer seed right after initialization and then never again.
2016-04-27 08:00:58 -07:00
Returns:
observation (object): the initial observation.
info (optional dictionary): a dictionary containing extra information, this is only returned if return_info is set to true
2016-04-27 08:00:58 -07:00
"""
# Initialize the RNG if the seed is manually passed
if seed is not None:
self._np_random, seed = seeding.np_random(seed)
2016-04-27 08:00:58 -07:00
@abstractmethod
2021-08-05 10:35:48 -04:00
def render(self, mode="human"):
2016-04-27 08:00:58 -07:00
"""Renders the environment.
The set of supported modes varies per environment. (And some
2022-02-13 01:39:03 +01:00
third-party environments may not support rendering at all.)
By convention, if mode is:
2016-04-27 08:00:58 -07:00
- human: render to the current display or terminal and
return nothing. Usually for human consumption.
- rgb_array: Return an numpy.ndarray with shape (x, y, 3),
representing RGB values for an x-by-y pixel image, suitable
for turning into a video.
- ansi: Return a string (str) or StringIO.StringIO containing a
terminal-style text representation. The text can include newlines
and ANSI escape sequences (e.g. for colors).
Note:
Make sure that your class's metadata 'render.modes' key includes
the list of supported modes. It's recommended to call super()
in implementations to use the functionality of this method.
Args:
mode (str): the mode to render with
Example:
class MyEnv(Env):
metadata = {'render.modes': ['human', 'rgb_array']}
def render(self, mode='human'):
if mode == 'rgb_array':
return np.array(...) # return RGB frame suitable for video
elif mode == 'human':
... # pop up a window and render
else:
super(MyEnv, self).render(mode=mode) # just raise an exception
2016-04-27 08:00:58 -07:00
"""
raise NotImplementedError
2016-04-27 08:00:58 -07:00
def close(self):
"""Override close in your subclass to perform any necessary cleanup.
Environments will automatically close() themselves when
garbage collected or when the program exits.
"""
pass
[WIP] add support for seeding environments (#135) * Make environments seedable * Fix monitor bugs - Set monitor_id before setting the infix. This was a bug that would yield incorrect results with multiple monitors. - Remove extra pid from stats recorder filename. This should be purely cosmetic. * Start uploading seeds in episode_batch * Fix _bigint_from_bytes for python3 * Set seed explicitly in random_agent * Pass through seed argument * Also pass through random state to spaces * Pass random state into the observation/action spaces * Make all _seed methods return the list of used seeds * Switch over to np.random where possible * Start hashing seeds, and also seed doom engine * Fixup seeding determinism in many cases * Seed before loading the ROM * Make seeding more Python3 friendly * Make the MuJoCo skipping a bit more forgiving * Remove debugging PDB calls * Make setInt argument into raw bytes * Validate and upload seeds * Skip box2d * Make seeds smaller, and change representation of seeds in upload * Handle long seeds * Fix RandomAgent example to be deterministic * Handle integer types correctly in Python2 and Python3 * Try caching pip * Try adding swap * Add df and free calls * Bump swap * Bump swap size * Try setting overcommit * Try other sysctls * Try fixing overcommit * Try just setting overcommit_memory=1 * Add explanatory comment * Add what's new section to readme * BUG: Mark ElevatorAction-ram-v0 as non-deterministic for now * Document seed * Move nondetermistic check into spec
2016-05-29 09:07:09 -07:00
def seed(self, seed=None):
"""Sets the seed for this env's random number generator(s).
Note:
Some environments use multiple pseudorandom number generators.
We want to capture all such seeds used in order to ensure that
there aren't accidental correlations between multiple generators.
Returns:
list<bigint>: Returns the list of seeds used in this env's random
number generators. The first value in the list should be the
"main" seed, or the value which a reproducer should pass to
'seed'. Often, the main seed equals the provided 'seed', but
this won't be true if seed=None, for example.
"""
Seeding update (#2422) * Ditch most of the seeding.py and replace np_random with the numpy default_rng. Let's see if tests pass * Updated a bunch of RNG calls from the RandomState API to Generator API * black; didn't expect that, did ya? * Undo a typo * blaaack * More typo fixes * Fixed setting/getting state in multidiscrete spaces * Fix typo, fix a test to work with the new sampling * Correctly (?) pass the randomly generated seed if np_random is called with None as seed * Convert the Discrete sample to a python int (as opposed to np.int64) * Remove some redundant imports * First version of the compatibility layer for old-style RNG. Mainly to trigger tests. * Removed redundant f-strings * Style fixes, removing unused imports * Try to make tests pass by removing atari from the dockerfile * Try to make tests pass by removing atari from the setup * Try to make tests pass by removing atari from the setup * Try to make tests pass by removing atari from the setup * First attempt at deprecating `env.seed` and supporting `env.reset(seed=seed)` instead. Tests should hopefully pass but throw up a million warnings. * black; didn't expect that, didya? * Rename the reset parameter in VecEnvs back to `seed` * Updated tests to use the new seeding method * Removed a bunch of old `seed` calls. Fixed a bug in AsyncVectorEnv * Stop Discrete envs from doing part of the setup (and using the randomness) in init (as opposed to reset) * Add explicit seed to wrappers reset * Remove an accidental return * Re-add some legacy functions with a warning. * Use deprecation instead of regular warnings for the newly deprecated methods/functions
2021-12-08 22:14:15 +01:00
deprecation(
"Function `env.seed(seed)` is marked as deprecated and will be removed in the future. "
"Please use `env.reset(seed=seed) instead."
)
self._np_random, seed = seeding.np_random(seed)
Seeding update (#2422) * Ditch most of the seeding.py and replace np_random with the numpy default_rng. Let's see if tests pass * Updated a bunch of RNG calls from the RandomState API to Generator API * black; didn't expect that, did ya? * Undo a typo * blaaack * More typo fixes * Fixed setting/getting state in multidiscrete spaces * Fix typo, fix a test to work with the new sampling * Correctly (?) pass the randomly generated seed if np_random is called with None as seed * Convert the Discrete sample to a python int (as opposed to np.int64) * Remove some redundant imports * First version of the compatibility layer for old-style RNG. Mainly to trigger tests. * Removed redundant f-strings * Style fixes, removing unused imports * Try to make tests pass by removing atari from the dockerfile * Try to make tests pass by removing atari from the setup * Try to make tests pass by removing atari from the setup * Try to make tests pass by removing atari from the setup * First attempt at deprecating `env.seed` and supporting `env.reset(seed=seed)` instead. Tests should hopefully pass but throw up a million warnings. * black; didn't expect that, didya? * Rename the reset parameter in VecEnvs back to `seed` * Updated tests to use the new seeding method * Removed a bunch of old `seed` calls. Fixed a bug in AsyncVectorEnv * Stop Discrete envs from doing part of the setup (and using the randomness) in init (as opposed to reset) * Add explicit seed to wrappers reset * Remove an accidental return * Re-add some legacy functions with a warning. * Use deprecation instead of regular warnings for the newly deprecated methods/functions
2021-12-08 22:14:15 +01:00
return [seed]
@property
def unwrapped(self) -> Env:
"""Completely unwrap this env.
2016-08-11 14:45:52 -07:00
Returns:
gym.Env: The base non-wrapped gym.Env instance
2016-08-11 14:45:52 -07:00
"""
2017-02-26 00:01:00 -08:00
return self
2016-08-11 14:45:52 -07:00
2016-04-27 08:00:58 -07:00
def __str__(self):
if self.spec is None:
return f"<{type(self).__name__} instance>"
else:
return f"<{type(self).__name__}<{self.spec.id}>>"
2016-04-27 08:00:58 -07:00
def __enter__(self):
2021-07-29 02:26:34 +02:00
"""Support with-statement for the environment."""
return self
def __exit__(self, *args):
2021-07-29 02:26:34 +02:00
"""Support with-statement for the environment."""
self.close()
# propagate exception
return False
2018-02-26 17:35:07 +01:00
class Wrapper(Env[ObsType, ActType]):
"""Wraps the environment to allow a modular transformation.
This class is the base class for all wrappers. The subclass could override
some methods to change the behavior of the original environment without touching the
original code.
.. note::
Don't forget to call ``super().__init__(env)`` if the subclass overrides :meth:`__init__`.
"""
2021-07-29 02:26:34 +02:00
def __init__(self, env: Env):
self.env = env
self._action_space: spaces.Space | None = None
self._observation_space: spaces.Space | None = None
self._reward_range: tuple[SupportsFloat, SupportsFloat] | None = None
self._metadata: dict | None = None
def __getattr__(self, name):
2021-07-29 02:26:34 +02:00
if name.startswith("_"):
raise AttributeError(f"attempted to get missing private attribute '{name}'")
return getattr(self.env, name)
@property
def spec(self):
return self.env.spec
@classmethod
def class_name(cls):
return cls.__name__
@property
def action_space(self) -> spaces.Space[ActType]:
if self._action_space is None:
return self.env.action_space
return self._action_space
@action_space.setter
def action_space(self, space):
self._action_space = space
@property
def observation_space(self) -> spaces.Space:
if self._observation_space is None:
return self.env.observation_space
return self._observation_space
@observation_space.setter
def observation_space(self, space):
self._observation_space = space
@property
def reward_range(self) -> tuple[SupportsFloat, SupportsFloat]:
if self._reward_range is None:
return self.env.reward_range
return self._reward_range
@reward_range.setter
def reward_range(self, value):
self._reward_range = value
@property
def metadata(self) -> dict:
if self._metadata is None:
return self.env.metadata
return self._metadata
@metadata.setter
def metadata(self, value):
self._metadata = value
def step(self, action: ActType) -> Tuple[ObsType, float, bool, dict]:
return self.env.step(action)
def reset(self, **kwargs) -> Union[ObsType, tuple[ObsType, dict]]:
return self.env.reset(**kwargs)
2016-08-11 14:45:52 -07:00
2021-07-29 02:26:34 +02:00
def render(self, mode="human", **kwargs):
return self.env.render(mode, **kwargs)
2016-08-11 14:45:52 -07:00
def close(self):
return self.env.close()
2016-08-11 14:45:52 -07:00
def seed(self, seed=None):
2016-08-11 14:45:52 -07:00
return self.env.seed(seed)
def __str__(self):
return f"<{type(self).__name__}{self.env}>"
def __repr__(self):
return str(self)
@property
def unwrapped(self) -> Env:
2017-02-26 00:01:00 -08:00
return self.env.unwrapped
class ObservationWrapper(Wrapper):
def reset(self, **kwargs):
if kwargs.get("return_info", False):
obs, info = self.env.reset(**kwargs)
return self.observation(obs), info
else:
return self.observation(self.env.reset(**kwargs))
def step(self, action):
observation, reward, done, info = self.env.step(action)
return self.observation(observation), reward, done, info
@abstractmethod
def observation(self, observation):
raise NotImplementedError
class RewardWrapper(Wrapper):
def reset(self, **kwargs):
return self.env.reset(**kwargs)
def step(self, action):
observation, reward, done, info = self.env.step(action)
return observation, self.reward(reward), done, info
@abstractmethod
def reward(self, reward):
raise NotImplementedError
class ActionWrapper(Wrapper):
def reset(self, **kwargs):
return self.env.reset(**kwargs)
def step(self, action):
return self.env.step(self.action(action))
@abstractmethod
def action(self, action):
raise NotImplementedError
@abstractmethod
2016-10-14 22:07:47 -07:00
def reverse_action(self, action):
2021-07-29 02:26:34 +02:00
raise NotImplementedError