def close(self):
"""Make sure to manually close, or else you'll leak the encoder process"""
if not self.enabled:
return
if self.encoder:
logger.debug('Closing video encoder: path=%s', self.path)
self.encoder.close()
self.encoder = None
else:
# No frames captured. Set metadata, and remove the empty output file.
os.remove(self.path)
if self.metadata is None:
self.metadata = {}
self.metadata['empty'] = True
# If broken, get rid of the output file, otherwise we'd leak it.
if self.broken:
logger.info(
'Cleaning up paths for broken video recorder: path=%s metadata_path=%s',
self.path, self.metadata_path)
# Might have crashed before even starting the output file, don't try to remove in that case.
if os.path.exists(self.path):
os.remove(self.path)
if self.metadata is None:
self.metadata = {}
self.metadata['broken'] = True
self.write_metadata()
def test_env_semantics(spec):
logger.warn("Skipping this test. Existing hashes were generated in a bad way")
return
with open(ROLLOUT_FILE) as data_file:
rollout_dict = json.load(data_file)
if spec.id not in rollout_dict:
if not spec.nondeterministic:
logger.warn("Rollout does not exist for {}, run generate_json.py to generate rollouts for new envs".format(spec.id))
return
logger.info("Testing rollout for {} environment...".format(spec.id))
observations_now, actions_now, rewards_now, dones_now = generate_rollout_hash(spec)
errors = []
if rollout_dict[spec.id]['observations'] != observations_now:
errors.append('Observations not equal for {} -- expected {} but got {}'.format(spec.id, rollout_dict[spec.id]['observations'], observations_now))
if rollout_dict[spec.id]['actions'] != actions_now:
errors.append('Actions not equal for {} -- expected {} but got {}'.format(spec.id, rollout_dict[spec.id]['actions'], actions_now))
if rollout_dict[spec.id]['rewards'] != rewards_now:
errors.append('Rewards not equal for {} -- expected {} but got {}'.format(spec.id, rollout_dict[spec.id]['rewards'], rewards_now))
if rollout_dict[spec.id]['dones'] != dones_now:
errors.append('Dones not equal for {} -- expected {} but got {}'.format(spec.id, rollout_dict[spec.id]['dones'], dones_now))
if len(errors):
for error in errors:
logger.warn(error)
raise ValueError(errors)
def clear_monitor_files(training_dir):
files = detect_monitor_files(training_dir)
if len(files) == 0:
return
logger.info(
'Clearing %d monitor files from previous run (because force=True was provided)',
len(files))
for file in files:
os.unlink(file)
def close(self):
"""Flush all monitor data to disk and close any open rending windows."""
if not self.enabled:
return
self.stats_recorder.close()
if self.video_recorder is not None:
self._close_video_recorder()
self._flush(force=True)
# Stop tracking this for autoclose
monitor_closer.unregister(self._monitor_id)
self.enabled = False
logger.info(
'''Finished writing results. You can upload them to the scoreboard via gym_wmgds.upload(%r)''',
self.directory)
def make(self, id, **kwargs):
if len(kwargs) > 0:
logger.info('Making new env: %s (%s)', id, kwargs)
else:
logger.info('Making new env: %s', id)
spec = self.spec(id)
env = spec.make(**kwargs)
# We used to have people override _reset/_step rather than
# reset/step. Set _gym_wmgds_disable_underscore_compat = True on
# your environment if you use these methods and don't want
# compatibility code to be invoked.
if hasattr(env, "_reset") and hasattr(env, "_step") and not getattr(
env, "_gym_wmgds_disable_underscore_compat", False):
patch_deprecated_methods(env)
if (env.spec.timestep_limit is not None) and not spec.tags.get('vnc'):
from gym_wmgds.wrappers.time_limit import TimeLimit
env = TimeLimit(env,
max_episode_steps=env.spec.max_episode_steps,
max_episode_seconds=env.spec.max_episode_seconds)
return env
def add_new_rollouts(spec_ids, overwrite):
environments = [
spec for spec in envs.registry.all() if spec._entry_point is not None
]
if spec_ids:
environments = [spec for spec in environments if spec.id in spec_ids]
assert len(environments) == len(spec_ids), "Some specs not found"
with open(ROLLOUT_FILE) as data_file:
rollout_dict = json.load(data_file)
modified = False
for spec in environments:
if not overwrite and spec.id in rollout_dict:
logger.debug("Rollout already exists for {}. Skipping.".format(
spec.id))
else:
modified = update_rollout_dict(spec, rollout_dict) or modified
if modified:
logger.info("Writing new rollout file to {}".format(ROLLOUT_FILE))
with open(ROLLOUT_FILE, "w") as outfile:
json.dump(rollout_dict, outfile, indent=2, sort_keys=True)
else:
logger.info("No modifications needed.")
def update_rollout_dict(spec, rollout_dict):
"""
Takes as input the environment spec for which the rollout is to be generated,
and the existing dictionary of rollouts. Returns True iff the dictionary was
modified.
"""
# Skip platform-dependent
if should_skip_env_spec_for_tests(spec):
logger.info("Skipping tests for {}".format(spec.id))
return False
# Skip environments that are nondeterministic
if spec.nondeterministic:
logger.info("Skipping tests for nondeterministic env {}".format(
spec.id))
return False
logger.info("Generating rollout for {}".format(spec.id))
try:
observations_hash, actions_hash, rewards_hash, dones_hash = generate_rollout_hash(
spec)
except:
# If running the env generates an exception, don't write to the rollout file
logger.warn(
"Exception {} thrown while generating rollout for {}. Rollout not added."
.format(sys.exc_info()[0], spec.id))
return False
rollout = {}
rollout['observations'] = observations_hash
rollout['actions'] = actions_hash
rollout['rewards'] = rewards_hash
rollout['dones'] = dones_hash
existing = rollout_dict.get(spec.id)
if existing:
differs = False
for key, new_hash in rollout.items():
differs = differs or existing[key] != new_hash
if not differs:
logger.debug("Hashes match with existing for {}".format(spec.id))
return False
else:
logger.warn("Got new hash for {}. Overwriting.".format(spec.id))
rollout_dict[spec.id] = rollout
return True
def __init__(self,
env,
path=None,
metadata=None,
enabled=True,
base_path=None):
modes = env.metadata.get('render.modes', [])
self._async = env.metadata.get('semantics.async')
self.enabled = enabled
# Don't bother setting anything else if not enabled
if not self.enabled:
return
self.ansi_mode = False
if 'rgb_array' not in modes:
if 'ansi' in modes:
self.ansi_mode = True
else:
logger.info(
'Disabling video recorder because {} neither supports video mode "rgb_array" nor "ansi".'
.format(env))
# Whoops, turns out we shouldn't be enabled after all
self.enabled = False
return
if path is not None and base_path is not None:
raise error.Error(
"You can pass at most one of `path` or `base_path`.")
self.last_frame = None
self.env = env
required_ext = '.json' if self.ansi_mode else '.mp4'
if path is None:
if base_path is not None:
# Base path given, append ext
path = base_path + required_ext
else:
# Otherwise, just generate a unique filename
with tempfile.NamedTemporaryFile(suffix=required_ext,
delete=False) as f:
path = f.name
self.path = path
path_base, actual_ext = os.path.splitext(self.path)
if actual_ext != required_ext:
hint = " HINT: The environment is text-only, therefore we're recording its text output in a structured JSON format." if self.ansi_mode else ''
raise error.Error(
"Invalid path given: {} -- must have file extension {}.{}".
format(self.path, required_ext, hint))
# Touch the file in any case, so we know it's present. (This
# corrects for platform platform differences. Using ffmpeg on
# OS X, the file is precreated, but not on Linux.
touch(path)
self.frames_per_sec = env.metadata.get('video.frames_per_second', 30)
self.encoder = None # lazily start the process
self.broken = False
# Dump metadata
self.metadata = metadata or {}
self.metadata[
'content_type'] = 'video/vnd.openai.ansivid' if self.ansi_mode else 'video/mp4'
self.metadata_path = '{}.meta.json'.format(path_base)
self.write_metadata()
logger.info('Starting new video recorder writing to %s', self.path)
self.empty = True
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_wmgds.").
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_wmgds.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_wmgds' 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)
def set_monitor_mode(self, mode):
logger.info(
"Setting the monitor mode is deprecated and will be removed soon")
self._set_mode(mode)
import sys
import argparse
from gym_wmgds.envs.tests.spec_list import should_skip_env_spec_for_tests
from gym_wmgds.envs.tests.test_envs_semantics import generate_rollout_hash, hash_object
DATA_DIR = os.path.join(os.path.dirname(__file__), os.pardir, 'gym_wmgds',
'envs', 'tests')
ROLLOUT_STEPS = 100
episodes = ROLLOUT_STEPS
steps = ROLLOUT_STEPS
ROLLOUT_FILE = os.path.join(DATA_DIR, 'rollout.json')
if not os.path.isfile(ROLLOUT_FILE):
logger.info("No rollout file found. Writing empty json file to {}".format(
ROLLOUT_FILE))
with open(ROLLOUT_FILE, "w") as outfile:
json.dump({}, outfile, indent=2)
def update_rollout_dict(spec, rollout_dict):
"""
Takes as input the environment spec for which the rollout is to be generated,
and the existing dictionary of rollouts. Returns True iff the dictionary was
modified.
"""
# Skip platform-dependent
if should_skip_env_spec_for_tests(spec):
logger.info("Skipping tests for {}".format(spec.id))
return False