def __init__(self, low, high, shape=None, dtype=np.float32):
assert dtype is not None, 'dtype must be explicitly provided. '
self.dtype = np.dtype(dtype)
if shape is None:
assert low.shape == high.shape, 'box dimension mismatch. '
self.shape = low.shape
self.low = low
self.high = high
else:
assert np.isscalar(low) and np.isscalar(high), 'box requires scalar bounds. '
self.shape = tuple(shape)
self.low = np.full(self.shape, low)
self.high = np.full(self.shape, high)
def _get_precision(dtype):
if np.issubdtype(dtype, np.floating):
return np.finfo(dtype).precision
else:
return np.inf
low_precision = _get_precision(self.low.dtype)
high_precision = _get_precision(self.high.dtype)
dtype_precision = _get_precision(self.dtype)
if min(low_precision, high_precision) > dtype_precision:
logger.warn("Box bound precision lowered by casting to {}".format(self.dtype))
self.low = self.low.astype(self.dtype)
self.high = self.high.astype(self.dtype)
# Boolean arrays which indicate the interval type for each coordinate
self.bounded_below = -np.inf < self.low
self.bounded_above = np.inf > self.high
super(Box, self).__init__(self.shape, self.dtype)
def _encode_image_frame(self, frame):
if not self.encoder:
self.encoder = ImageEncoder(self.path, frame.shape, self.frames_per_sec)
self.metadata['encoder_version'] = self.encoder.version_info
try:
self.encoder.capture_frame(frame)
except error.InvalidFrame as e:
logger.warn('Tried to pass invalid video frame, marking as broken: %s', e)
self.broken = True
else:
self.empty = False
def close_extras(self, timeout=None, terminate=False):
"""
Parameters
----------
timeout : int or float, optional
Number of seconds before the call to `close` times out. If `None`,
the call to `close` never times out. If the call to `close` times
out, then all processes are terminated.
terminate : bool (default: `False`)
If `True`, then the `close` operation is forced and all processes
are terminated.
"""
timeout = 0 if terminate else timeout
try:
if self._state != AsyncState.DEFAULT:
logger.warn('Calling `close` while waiting for a pending '
'call to `{0}` to complete.'.format(
self._state.value))
function = getattr(self, '{0}_wait'.format(self._state.value))
function(timeout)
except mp.TimeoutError:
terminate = True
if terminate:
for process in self.processes:
if process.is_alive():
process.terminate()
else:
for pipe in self.parent_pipes:
if (pipe is not None) and (not pipe.closed):
pipe.send(('close', None))
for pipe in self.parent_pipes:
if (pipe is not None) and (not pipe.closed):
pipe.recv()
for pipe in self.parent_pipes:
if pipe is not None:
pipe.close()
for process in self.processes:
process.join()
def capture_frame(self):
"""Render the given `env` and add the resulting frame to the video."""
if not self.functional: return
logger.debug('Capturing video frame: path=%s', self.path)
render_mode = 'ansi' if self.ansi_mode else 'rgb_array'
frame = self.env.render(mode=render_mode)
if frame is None:
if self._async:
return
else:
# Indicates a bug in the environment: don't want to raise
# an error here.
logger.warn('Env returned None on render(). Disabling further rendering for video recorder by marking as disabled: path=%s metadata_path=%s', self.path, self.metadata_path)
self.broken = True
else:
self.last_frame = frame
if self.ansi_mode:
self._encode_ansi_frame(frame)
else:
self._encode_image_frame(frame)
def patch_deprecated_methods(env):
"""
Methods renamed from '_method' to 'method', render() no longer has 'close' parameter, close is a separate method.
For backward compatibility, this makes it possible to work with unmodified environments.
"""
global warn_once
if warn_once:
logger.warn(
"Environment '%s' has deprecated methods '_step' and '_reset' rather than 'step' and 'reset'. Compatibility code invoked. Set _gym_disable_underscore_compat = True to disable this behavior."
% str(type(env)))
warn_once = False
env.reset = env._reset
env.step = env._step
env.seed = env._seed
def render(mode):
return env._render(mode, close=False)
def close():
env._render("human", close=True)
env.render = render
env.close = close
def __init__(self,
env_fns,
observation_space=None,
action_space=None,
shared_memory=True,
copy=True,
context=None,
daemon=True,
worker=None):
try:
ctx = mp.get_context(context)
except AttributeError:
logger.warn('Context switching for `multiprocessing` is not '
'available in Python 2. Using the default context.')
ctx = mp
self.env_fns = env_fns
self.shared_memory = shared_memory
self.copy = copy
if (observation_space is None) or (action_space is None):
dummy_env = env_fns[0]()
observation_space = observation_space or dummy_env.observation_space
action_space = action_space or dummy_env.action_space
dummy_env.close()
del dummy_env
super(AsyncVectorEnv,
self).__init__(num_envs=len(env_fns),
observation_space=observation_space,
action_space=action_space)
if self.shared_memory:
_obs_buffer = create_shared_memory(self.single_observation_space,
n=self.num_envs,
ctx=ctx)
self.observations = read_from_shared_memory(
_obs_buffer, self.single_observation_space, n=self.num_envs)
else:
_obs_buffer = None
self.observations = create_empty_array(
self.single_observation_space, n=self.num_envs, fn=np.zeros)
self.parent_pipes, self.processes = [], []
self.error_queue = ctx.Queue()
target = _worker_shared_memory if self.shared_memory else _worker
target = worker or target
with clear_mpi_env_vars():
for idx, env_fn in enumerate(self.env_fns):
parent_pipe, child_pipe = ctx.Pipe()
process = ctx.Process(
target=target,
name='Worker<{0}>-{1}'.format(type(self).__name__, idx),
args=(idx, CloudpickleWrapper(env_fn), child_pipe,
parent_pipe, _obs_buffer, self.error_queue))
self.parent_pipes.append(parent_pipe)
self.processes.append(process)
process.daemon = daemon
process.start()
child_pipe.close()
self._state = AsyncState.DEFAULT
self._check_observation_spaces()
import gym_open_ai
import pygame
import matplotlib
import argparse
from gym_open_ai import logger
try:
matplotlib.use('TkAgg')
import matplotlib.pyplot as plt
except ImportError as e:
logger.warn(
'failed to set matplotlib backend, plotting will not work: %s' %
str(e))
plt = None
from collections import deque
from pygame.locals import VIDEORESIZE
def display_arr(screen, arr, video_size, transpose):
arr_min, arr_max = arr.min(), arr.max()
arr = 255.0 * (arr - arr_min) / (arr_max - arr_min)
pyg_img = pygame.surfarray.make_surface(
arr.swapaxes(0, 1) if transpose else arr)
pyg_img = pygame.transform.scale(pyg_img, video_size)
screen.blit(pyg_img, (0, 0))
def play(env,
transpose=True,
fps=30,
zoom=None,
def _start(self,
directory,
video_callable=None,
force=False,
resume=False,
write_upon_reset=False,
uid=None,
mode=None):
"""Start monitoring.
Args:
directory (str): A per-training run directory where to record stats.
video_callable (Optional[function, False]): function that takes in the index of the episode and outputs a boolean, indicating whether we should record a video on this episode. The default (for video_callable is None) is to take perfect cubes, capped at 1000. False disables video recording.
force (bool): Clear out existing training data from this directory (by deleting every file prefixed with "openaigym.").
resume (bool): Retain the training data already in this directory, which will be merged with our new data
write_upon_reset (bool): Write the manifest file on each reset. (This is currently a JSON file, so writing it is somewhat expensive.)
uid (Optional[str]): A unique id used as part of the suffix for the file. By default, uses os.getpid().
mode (['evaluation', 'training']): Whether this is an evaluation or training episode.
"""
if self.env.spec is None:
logger.warn(
"Trying to monitor an environment which has no 'spec' set. This usually means you did not create it via 'gym_open_ai.make', and is recommended only for advanced users."
)
env_id = '(unknown)'
else:
env_id = self.env.spec.id
if not os.path.exists(directory):
logger.info('Creating monitor directory %s', directory)
if six.PY3:
os.makedirs(directory, exist_ok=True)
else:
os.makedirs(directory)
if video_callable is None:
video_callable = capped_cubic_video_schedule
elif video_callable == False:
video_callable = disable_videos
elif not callable(video_callable):
raise error.Error(
'You must provide a function, None, or False for video_callable, not {}: {}'
.format(type(video_callable), video_callable))
self.video_callable = video_callable
# Check on whether we need to clear anything
if force:
clear_monitor_files(directory)
elif not resume:
training_manifests = detect_training_manifests(directory)
if len(training_manifests) > 0:
raise error.Error(
'''Trying to write to monitor directory {} with existing monitor files: {}.
You should use a unique directory for each training run, or use 'force=True' to automatically clear previous monitor files.'''
.format(directory, ', '.join(training_manifests[:5])))
self._monitor_id = monitor_closer.register(self)
self.enabled = True
self.directory = os.path.abspath(directory)
# We use the 'openai-gym_open_ai' prefix to determine if a file is
# ours
self.file_prefix = FILE_PREFIX
self.file_infix = '{}.{}'.format(self._monitor_id,
uid if uid else os.getpid())
self.stats_recorder = stats_recorder.StatsRecorder(
directory,
'{}.episode_batch.{}'.format(self.file_prefix, self.file_infix),
autoreset=self.env_semantics_autoreset,
env_id=env_id)
if not os.path.exists(directory): os.mkdir(directory)
self.write_upon_reset = write_upon_reset
if mode is not None:
self._set_mode(mode)