def run_task(vv, log_dir=None, exp_name=None):
# Load environment
radius = vv['radius']
target_velocity = vv['target_velocity']
env = normalize(CircleEnv(radius, target_velocity))
# Save variant information for comparison plots
variant_file = logger.get_snapshot_dir() + '/variant.json'
logger.log_variant(variant_file, vv)
# Train policy using TRPO
policy = GaussianMLPPolicy(
env_spec=env.spec,
hidden_sizes=(32, 32)
)
baseline = LinearFeatureBaseline(env_spec=env.spec)
algo = TRPO(
env=env,
policy=policy,
baseline=baseline,
batch_size=1000,
max_path_length=env.horizon,
n_itr=1000,
discount=0.99,
step_size=0.01,
plot=False,
)
algo.train()
def setup_logger(
exp_prefix=None,
exp_count=0,
seed=0,
variant=None,
log_dir=None,
text_log_file="debug.log",
variant_log_file="variant.json",
tabular_log_file="progress.csv",
snapshot_mode="last",
log_tabular_only=False,
snapshot_gap=1,
):
"""
Set up logger to have some reasonable default settings.
:param exp_prefix:
:param exp_count:
:param seed: Experiment seed
:param variant:
:param log_dir:
:param text_log_file:
:param variant_log_file:
:param tabular_log_file:
:param snapshot_mode:
:param log_tabular_only:
:param snapshot_gap:
:return:
"""
if log_dir is None:
assert exp_prefix is not None
log_dir = create_log_dir(exp_prefix, exp_count=exp_count, seed=seed)
tabular_log_path = osp.join(log_dir, tabular_log_file)
text_log_path = osp.join(log_dir, text_log_file)
if variant is not None:
variant_log_path = osp.join(log_dir, variant_log_file)
logger.log_variant(variant_log_path, variant)
logger.add_text_output(text_log_path)
logger.add_tabular_output(tabular_log_path)
logger.set_snapshot_dir(log_dir)
logger.set_snapshot_mode(snapshot_mode)
logger.set_snapshot_gap(snapshot_gap)
logger.set_log_tabular_only(log_tabular_only)
def run_experiment(argv):
# e2crawfo: These imports, in this order, were necessary for fixing issues on cedar.
import rllab.mujoco_py.mjlib
import tensorflow
default_log_dir = config.LOG_DIR
now = datetime.datetime.now(dateutil.tz.tzlocal())
# avoid name clashes when running distributed jobs
rand_id = str(uuid.uuid4())[:5]
timestamp = now.strftime('%Y_%m_%d_%H_%M_%S_%f_%Z')
default_exp_name = 'experiment_%s_%s' % (timestamp, rand_id)
parser = argparse.ArgumentParser()
parser.add_argument(
'--n_parallel',
type=int,
default=1,
help=
'Number of parallel workers to perform rollouts. 0 => don\'t start any workers'
)
parser.add_argument('--exp_name',
type=str,
default=default_exp_name,
help='Name of the experiment.')
parser.add_argument('--log_dir',
type=str,
default=None,
help='Path to save the log and iteration snapshot.')
parser.add_argument('--snapshot_mode',
type=str,
default='all',
help='Mode to save the snapshot. Can be either "all" '
'(all iterations will be saved), "last" (only '
'the last iteration will be saved), or "none" '
'(do not save snapshots)')
parser.add_argument('--snapshot_gap',
type=int,
default=1,
help='Gap between snapshot iterations.')
parser.add_argument('--tabular_log_file',
type=str,
default='progress.csv',
help='Name of the tabular log file (in csv).')
parser.add_argument('--text_log_file',
type=str,
default='debug.log',
help='Name of the text log file (in pure text).')
parser.add_argument('--params_log_file',
type=str,
default='params.json',
help='Name of the parameter log file (in json).')
parser.add_argument('--variant_log_file',
type=str,
default='variant.json',
help='Name of the variant log file (in json).')
parser.add_argument(
'--resume_from',
type=str,
default=None,
help='Name of the pickle file to resume experiment from.')
parser.add_argument('--plot',
type=ast.literal_eval,
default=False,
help='Whether to plot the iteration results')
parser.add_argument(
'--log_tabular_only',
type=ast.literal_eval,
default=False,
help=
'Whether to only print the tabular log information (in a horizontal format)'
)
parser.add_argument('--seed', type=int, help='Random seed for numpy')
parser.add_argument('--args_data',
type=str,
help='Pickled data for stub objects')
parser.add_argument('--variant_data',
type=str,
help='Pickled data for variant configuration')
parser.add_argument('--use_cloudpickle',
type=ast.literal_eval,
default=False)
args = parser.parse_args(argv[1:])
if args.seed is not None:
set_seed(args.seed)
if args.n_parallel > 0:
from rllab.sampler import parallel_sampler
parallel_sampler.initialize(n_parallel=args.n_parallel)
if args.seed is not None:
parallel_sampler.set_seed(args.seed)
if args.plot:
from rllab.plotter import plotter
plotter.init_worker()
if args.log_dir is None:
log_dir = osp.join(default_log_dir, args.exp_name)
else:
log_dir = args.log_dir
tabular_log_file = osp.join(log_dir, args.tabular_log_file)
text_log_file = osp.join(log_dir, args.text_log_file)
params_log_file = osp.join(log_dir, args.params_log_file)
if args.variant_data is not None:
variant_data = pickle.loads(base64.b64decode(args.variant_data))
variant_log_file = osp.join(log_dir, args.variant_log_file)
logger.log_variant(variant_log_file, variant_data)
else:
variant_data = None
if not args.use_cloudpickle:
logger.log_parameters_lite(params_log_file, args)
logger.add_text_output(text_log_file)
logger.add_tabular_output(tabular_log_file)
prev_snapshot_dir = logger.get_snapshot_dir()
prev_mode = logger.get_snapshot_mode()
logger.set_snapshot_dir(log_dir)
logger.set_tf_summary_dir(osp.join(log_dir, "tf_summary"))
logger.set_snapshot_mode(args.snapshot_mode)
logger.set_snapshot_gap(args.snapshot_gap)
logger.set_log_tabular_only(args.log_tabular_only)
logger.push_prefix("[%s] " % args.exp_name)
if args.resume_from is not None:
data = joblib.load(args.resume_from)
assert 'algo' in data
algo = data['algo']
maybe_iter = algo.train()
if is_iterable(maybe_iter):
for _ in maybe_iter:
pass
else:
# read from stdin
if args.use_cloudpickle:
import cloudpickle
method_call = cloudpickle.loads(base64.b64decode(args.args_data))
method_call(variant_data)
else:
data = pickle.loads(base64.b64decode(args.args_data))
maybe_iter = concretize(data)
if is_iterable(maybe_iter):
for _ in maybe_iter:
pass
logger.set_snapshot_mode(prev_mode)
logger.set_snapshot_dir(prev_snapshot_dir)
logger.remove_tabular_output(tabular_log_file)
logger.remove_text_output(text_log_file)
logger.pop_prefix()
def run_experiment(argv):
default_log_dir = config.LOG_DIR
now = datetime.datetime.now(dateutil.tz.tzlocal())
# avoid name clashes when running distributed jobs
rand_id = str(uuid.uuid4())[:5]
timestamp = now.strftime('%Y_%m_%d_%H_%M_%S_%f_%Z')
default_exp_name = 'experiment_%s_%s' % (timestamp, rand_id)
parser = argparse.ArgumentParser()
parser.add_argument('--n_parallel', type=int, default=1,
help='Number of parallel workers to perform rollouts. 0 => don\'t start any workers')
parser.add_argument(
'--exp_name', type=str, default=default_exp_name, help='Name of the experiment.')
parser.add_argument('--log_dir', type=str, default=None,
help='Path to save the log and iteration snapshot.')
parser.add_argument('--snapshot_mode', type=str, default='all',
help='Mode to save the snapshot. Can be either "all" '
'(all iterations will be saved), "last" (only '
'the last iteration will be saved), "gap" (every'
'`snapshot_gap` iterations are saved), or "none" '
'(do not save snapshots)')
parser.add_argument('--snapshot_gap', type=int, default=1,
help='Gap between snapshot iterations.')
parser.add_argument('--tabular_log_file', type=str, default='progress.csv',
help='Name of the tabular log file (in csv).')
parser.add_argument('--text_log_file', type=str, default='debug.log',
help='Name of the text log file (in pure text).')
parser.add_argument('--params_log_file', type=str, default='params.json',
help='Name of the parameter log file (in json).')
parser.add_argument('--variant_log_file', type=str, default='variant.json',
help='Name of the variant log file (in json).')
parser.add_argument('--resume_from', type=str, default=None,
help='Name of the pickle file to resume experiment from.')
parser.add_argument('--plot', type=ast.literal_eval, default=False,
help='Whether to plot the iteration results')
parser.add_argument('--log_tabular_only', type=ast.literal_eval, default=False,
help='Whether to only print the tabular log information (in a horizontal format)')
parser.add_argument('--seed', type=int,
help='Random seed for numpy')
parser.add_argument('--args_data', type=str,
help='Pickled data for stub objects')
parser.add_argument('--variant_data', type=str,
help='Pickled data for variant configuration')
parser.add_argument('--use_cloudpickle', type=ast.literal_eval, default=False)
args = parser.parse_args(argv[1:])
if args.seed is not None:
set_seed(args.seed)
if args.n_parallel > 0:
from rllab.sampler import parallel_sampler
parallel_sampler.initialize(n_parallel=args.n_parallel)
if args.seed is not None:
parallel_sampler.set_seed(args.seed)
if args.plot:
from rllab.plotter import plotter
plotter.init_worker()
if args.log_dir is None:
log_dir = osp.join(default_log_dir, args.exp_name)
else:
log_dir = args.log_dir
tabular_log_file = osp.join(log_dir, args.tabular_log_file)
text_log_file = osp.join(log_dir, args.text_log_file)
params_log_file = osp.join(log_dir, args.params_log_file)
if args.variant_data is not None:
variant_data = pickle.loads(base64.b64decode(args.variant_data))
variant_log_file = osp.join(log_dir, args.variant_log_file)
logger.log_variant(variant_log_file, variant_data)
else:
variant_data = None
if not args.use_cloudpickle:
logger.log_parameters_lite(params_log_file, args)
logger.add_text_output(text_log_file)
logger.add_tabular_output(tabular_log_file)
prev_snapshot_dir = logger.get_snapshot_dir()
prev_mode = logger.get_snapshot_mode()
logger.set_snapshot_dir(log_dir)
logger.set_snapshot_mode(args.snapshot_mode)
logger.set_snapshot_gap(args.snapshot_gap)
logger.set_log_tabular_only(args.log_tabular_only)
logger.push_prefix("[%s] " % args.exp_name)
if args.resume_from is not None:
data = joblib.load(args.resume_from)
assert 'algo' in data
algo = data['algo']
algo.train()
else:
# read from stdin
if args.use_cloudpickle:
import cloudpickle
method_call = cloudpickle.loads(base64.b64decode(args.args_data))
method_call(variant_data)
else:
data = pickle.loads(base64.b64decode(args.args_data))
maybe_iter = concretize(data)
if is_iterable(maybe_iter):
for _ in maybe_iter:
pass
logger.set_snapshot_mode(prev_mode)
logger.set_snapshot_dir(prev_snapshot_dir)
logger.remove_tabular_output(tabular_log_file)
logger.remove_text_output(text_log_file)
logger.pop_prefix()
def run_task(vv, log_dir=None, exp_name=None):
global policy
global baseline
trpo_stepsize = 0.01
trpo_subsample_factor = 0.2
# Check if variant is available
if vv['model_type'] not in ['BrushTireModel', 'LinearTireModel']:
raise ValueError('Unrecognized model type for simulating robot')
if vv['robot_type'] not in ['MRZR', 'RCCar']:
raise ValueError('Unrecognized robot type')
# Load environment
if not vv['use_ros']:
env = CircleEnv(target_velocity=vv['target_velocity'],
radius=vv['radius'],
dt=vv['dt'],
model_type=vv['model_type'],
robot_type=vv['robot_type'])
else:
from aa_simulation.envs.circle.circle_env_ros import CircleEnvROS
env = CircleEnvROS(target_velocity=vv['target_velocity'],
radius=vv['radius'],
dt=vv['dt'],
model_type=vv['model_type'],
robot_type=vv['robot_type'])
# Save variant information for comparison plots
variant_file = logger.get_snapshot_dir() + '/variant.json'
logger.log_variant(variant_file, vv)
# Set variance for each action component separately for exploration
# Note: We set the variance manually because we are not scaling our
# action space during training.
init_std_speed = vv['target_velocity'] / 4
init_std_steer = np.pi / 6
init_std = [init_std_speed, init_std_steer]
# Build policy and baseline networks
# Note: Mean of policy network set to analytically computed values for
# faster training (rough estimates for RL to fine-tune).
if policy is None or baseline is None:
wheelbase = 0.257
target_velocity = vv['target_velocity']
target_steering = np.arctan(wheelbase / vv['radius']) # CCW
output_mean = np.array([target_velocity, target_steering])
hidden_sizes = (32, 32)
# In mean network, allow output b values to dominate final output
# value by constraining the magnitude of the output W matrix. This is
# to allow faster learning. These numbers are arbitrarily chosen.
W_gain = min(vv['target_velocity'] / 5, np.pi / 15)
mean_network = MLP(input_shape=(env.spec.observation_space.flat_dim, ),
output_dim=env.spec.action_space.flat_dim,
hidden_sizes=hidden_sizes,
hidden_nonlinearity=LN.tanh,
output_nonlinearity=None,
output_W_init=LI.GlorotUniform(gain=W_gain),
output_b_init=output_mean)
policy = GaussianMLPPolicy(env_spec=env.spec,
hidden_sizes=hidden_sizes,
init_std=init_std,
mean_network=mean_network)
baseline = LinearFeatureBaseline(env_spec=env.spec,
target_key='returns')
# Reset variance to re-enable exploration when using pre-trained networks
else:
policy._l_log_std = ParamLayer(
policy._mean_network.input_layer,
num_units=env.spec.action_space.flat_dim,
param=LI.Constant(np.log(init_std)),
name='output_log_std',
trainable=True)
obs_var = policy._mean_network.input_layer.input_var
mean_var, log_std_var = L.get_output(
[policy._l_mean, policy._l_log_std])
policy._log_std_var = log_std_var
LasagnePowered.__init__(policy, [policy._l_mean, policy._l_log_std])
policy._f_dist = ext.compile_function(inputs=[obs_var],
outputs=[mean_var, log_std_var])
safety_baseline = LinearFeatureBaseline(env_spec=env.spec,
target_key='safety_returns')
safety_constraint = CircleSafetyConstraint(max_value=1.0,
eps=vv['eps'],
baseline=safety_baseline)
if vv['algo'] == 'TRPO':
algo = TRPO(
env=env,
policy=policy,
baseline=baseline,
batch_size=600,
max_path_length=env.horizon,
n_itr=600,
discount=0.99,
step_size=trpo_stepsize,
plot=False,
)
else:
algo = CPO(env=env,
policy=policy,
baseline=baseline,
safety_constraint=safety_constraint,
batch_size=600,
max_path_length=env.horizon,
n_itr=600,
discount=0.99,
step_size=trpo_stepsize,
gae_lambda=0.95,
safety_gae_lambda=1,
optimizer_args={'subsample_factor': trpo_subsample_factor},
plot=False)
algo.train()
def run_experiment(argv):
default_log_dir = config.LOG_DIR
now = datetime.datetime.now(dateutil.tz.tzlocal())
# avoid name clashes when running distributed jobs
rand_id = str(uuid.uuid4())[:5]
timestamp = now.strftime('%Y_%m_%d_%H_%M_%S_%f_%Z')
default_exp_name = 'experiment_%s_%s' % (timestamp, rand_id)
parser = argparse.ArgumentParser()
parser.add_argument(
'--n_parallel',
type=int,
default=1,
help=
'Number of parallel workers to perform rollouts. 0 => don\'t start any workers'
)
parser.add_argument('--exp_name',
type=str,
default=default_exp_name,
help='Name of the experiment.')
parser.add_argument('--log_dir',
type=str,
default=None,
help='Path to save the log and iteration snapshot.')
parser.add_argument('--snapshot_mode',
type=str,
default='all',
help='Mode to save the snapshot. Can be either "all" '
'(all iterations will be saved), "last" (only '
'the last iteration will be saved), "gap" (every'
'`snapshot_gap` iterations are saved), or "none" '
'(do not save snapshots)')
parser.add_argument('--snapshot_gap',
type=int,
default=1,
help='Gap between snapshot iterations.')
parser.add_argument('--tabular_log_file',
type=str,
default='progress.csv',
help='Name of the tabular log file (in csv).')
parser.add_argument('--text_log_file',
type=str,
default='debug.log',
help='Name of the text log file (in pure text).')
parser.add_argument('--params_log_file',
type=str,
default='params.json',
help='Name of the parameter log file (in json).')
parser.add_argument('--variant_log_file',
type=str,
default='variant.json',
help='Name of the variant log file (in json).')
parser.add_argument(
'--resume_from',
type=str,
default=None,
help='Name of the pickle file to resume experiment from.')
parser.add_argument('--plot',
type=ast.literal_eval,
default=False,
help='Whether to plot the iteration results')
parser.add_argument(
'--log_tabular_only',
type=ast.literal_eval,
default=False,
help=
'Whether to only print the tabular log information (in a horizontal format)'
)
parser.add_argument('--seed', type=int, help='Random seed for numpy')
parser.add_argument('--args_data',
type=str,
help='Pickled data for stub objects')
parser.add_argument('--variant_data',
type=str,
help='Pickled data for variant configuration')
parser.add_argument('--use_cloudpickle',
type=ast.literal_eval,
default=False)
args = parser.parse_args(argv[1:])
if args.seed is not None:
set_seed(args.seed)
if args.n_parallel > 0:
from rllab.sampler import parallel_sampler
parallel_sampler.initialize(n_parallel=args.n_parallel)
if args.seed is not None:
parallel_sampler.set_seed(args.seed)
if args.plot:
from rllab.plotter import plotter
plotter.init_worker()
if args.log_dir is None:
log_dir = osp.join(default_log_dir, args.exp_name)
else:
log_dir = args.log_dir
tabular_log_file = osp.join(log_dir, args.tabular_log_file)
text_log_file = osp.join(log_dir, args.text_log_file)
params_log_file = osp.join(log_dir, args.params_log_file)
if args.variant_data is not None:
variant_data = pickle.loads(base64.b64decode(args.variant_data))
variant_log_file = osp.join(log_dir, args.variant_log_file)
logger.log_variant(variant_log_file, variant_data)
else:
variant_data = None
if not args.use_cloudpickle:
logger.log_parameters_lite(params_log_file, args)
logger.add_text_output(text_log_file)
logger.add_tabular_output(tabular_log_file)
prev_snapshot_dir = logger.get_snapshot_dir()
prev_mode = logger.get_snapshot_mode()
logger.set_snapshot_dir(log_dir)
logger.set_snapshot_mode(args.snapshot_mode)
logger.set_snapshot_gap(args.snapshot_gap)
logger.set_log_tabular_only(args.log_tabular_only)
logger.push_prefix("[%s] " % args.exp_name)
#variant_data is the variant dictionary sent from trpoTests_ExpLite
if (args.resume_from is not None) and (
'&|&' in args.resume_from
): #separate string on &|& to get iters and file location
vals = args.resume_from.split(
'&|&') #dirRes | numItrs to go | new batchSize
dirRes = vals[0]
numItrs = int(vals[1])
if (len(vals) > 2):
batchSize = int(vals[2])
print("resuming from :{}".format(dirRes))
data = joblib.load(dirRes)
#data is dict : 'baseline', 'algo', 'itr', 'policy', 'env'
assert 'algo' in data
algo = data['algo']
assert 'policy' in data
pol = data['policy']
bl = data['baseline']
oldBatchSize = algo.batch_size
algo.n_itr = numItrs
if (len(vals) > 2):
algo.batch_size = batchSize
print(
'algo iters : {} cur iter :{} oldBatchSize : {} newBatchSize : {}'
.format(algo.n_itr, algo.current_itr, oldBatchSize,
algo.batch_size))
else:
print('algo iters : {} cur iter :{} '.format(
algo.n_itr, algo.current_itr))
algo.train()
else:
print('Not resuming - building new exp')
# read from stdin
if args.use_cloudpickle: #set to use cloudpickle
import cloudpickle
method_call = cloudpickle.loads(base64.b64decode(args.args_data))
method_call(variant_data)
else:
print('not use cloud pickle')
data = pickle.loads(base64.b64decode(args.args_data))
maybe_iter = concretize(data)
if is_iterable(maybe_iter):
for _ in maybe_iter:
pass
logger.set_snapshot_mode(prev_mode)
logger.set_snapshot_dir(prev_snapshot_dir)
logger.remove_tabular_output(tabular_log_file)
logger.remove_text_output(text_log_file)
logger.pop_prefix()
def run_experiment(argv):
default_log_dir = config.LOG_DIR
now = datetime.datetime.now(dateutil.tz.tzlocal())
# avoid name clashes when running distributed jobs
rand_id = str(uuid.uuid4())[:5]
timestamp = now.strftime('%Y_%m_%d_%H_%M_%S_%f_%Z')
default_exp_name = 'experiment_%s_%s' % (timestamp, rand_id)
parser = argparse.ArgumentParser()
parser.add_argument(
'--n_parallel',
type=int,
default=1,
help=
'Number of parallel workers to perform rollouts. 0 => don\'t start any workers'
)
parser.add_argument('--exp_name',
type=str,
default=default_exp_name,
help='Name of the experiment.')
parser.add_argument('--log_dir',
type=str,
default=None,
help='Path to save the log and iteration snapshot.')
parser.add_argument('--snapshot_mode',
type=str,
default='all',
help='Mode to save the snapshot. Can be either "all" '
'(all iterations will be saved), "last" (only '
'the last iteration will be saved), "gap" (every'
'`snapshot_gap` iterations are saved), or "none" '
'(do not save snapshots)')
parser.add_argument('--snapshot_gap',
type=int,
default=1,
help='Gap between snapshot iterations.')
parser.add_argument('--tabular_log_file',
type=str,
default='progress.csv',
help='Name of the tabular log file (in csv).')
parser.add_argument('--text_log_file',
type=str,
default='debug.log',
help='Name of the text log file (in pure text).')
parser.add_argument('--tensorboard_log_dir',
type=str,
default='tb',
help='Name of the folder for tensorboard_summary.')
parser.add_argument(
'--tensorboard_step_key',
type=str,
default=None,
help=
'Name of the step key in log data which shows the step in tensorboard_summary.'
)
parser.add_argument('--params_log_file',
type=str,
default='params.json',
help='Name of the parameter log file (in json).')
parser.add_argument('--variant_log_file',
type=str,
default='variant.json',
help='Name of the variant log file (in json).')
parser.add_argument(
'--resume_from',
type=str,
default=None,
help='Name of the pickle file to resume experiment from.')
parser.add_argument('--plot',
type=ast.literal_eval,
default=False,
help='Whether to plot the iteration results')
parser.add_argument(
'--log_tabular_only',
type=ast.literal_eval,
default=False,
help=
'Whether to only print the tabular log information (in a horizontal format)'
)
parser.add_argument('--seed', type=int, help='Random seed for numpy')
parser.add_argument('--args_data',
type=str,
help='Pickled data for stub objects')
parser.add_argument('--variant_data',
type=str,
help='Pickled data for variant configuration')
parser.add_argument('--use_cloudpickle',
type=ast.literal_eval,
default=False)
parser.add_argument('--checkpoint_dir',
type=str,
default='checkpoint',
help='Name of the folder for checkpoints.')
parser.add_argument('--obs_dir',
type=str,
default='obs',
help='Name of the folder for original observations.')
args = parser.parse_args(argv[1:])
if args.seed is not None:
set_seed(args.seed)
if args.n_parallel > 0:
from rllab.sampler import parallel_sampler
parallel_sampler.initialize(n_parallel=args.n_parallel)
if args.seed is not None:
parallel_sampler.set_seed(args.seed)
if args.plot:
from rllab.plotter import plotter
plotter.init_worker()
if args.log_dir is None:
log_dir = osp.join(default_log_dir, args.exp_name)
else:
log_dir = args.log_dir
tabular_log_file = osp.join(log_dir, args.tabular_log_file)
text_log_file = osp.join(log_dir, args.text_log_file)
params_log_file = osp.join(log_dir, args.params_log_file)
tensorboard_log_dir = osp.join(log_dir, args.tensorboard_log_dir)
checkpoint_dir = osp.join(log_dir, args.checkpoint_dir)
obs_dir = osp.join(log_dir, args.obs_dir)
if args.variant_data is not None:
variant_data = pickle.loads(base64.b64decode(args.variant_data))
variant_log_file = osp.join(log_dir, args.variant_log_file)
logger.log_variant(variant_log_file, variant_data)
else:
variant_data = None
if not args.use_cloudpickle:
logger.log_parameters_lite(params_log_file, args)
logger.add_text_output(text_log_file)
logger.add_tabular_output(tabular_log_file)
logger.set_tensorboard_dir(tensorboard_log_dir)
logger.set_checkpoint_dir(checkpoint_dir)
logger.set_obs_dir(obs_dir)
prev_snapshot_dir = logger.get_snapshot_dir()
prev_mode = logger.get_snapshot_mode()
logger.set_snapshot_dir(log_dir)
logger.set_snapshot_mode(args.snapshot_mode)
logger.set_snapshot_gap(args.snapshot_gap)
logger.set_log_tabular_only(args.log_tabular_only)
logger.set_tensorboard_step_key(args.tensorboard_step_key)
logger.push_prefix("[%s] " % args.exp_name)
git_commit = get_git_commit_hash()
logger.log('Git commit: {}'.format(git_commit))
git_diff_file_path = osp.join(log_dir,
'git_diff_{}.patch'.format(git_commit))
save_git_diff_to_file(git_diff_file_path)
logger.log('hostname: {}, pid: {}, tmux session: {}'.format(
socket.gethostname(), os.getpid(), get_tmux_session_name()))
if args.resume_from is not None:
data = joblib.load(args.resume_from)
assert 'algo' in data
algo = data['algo']
algo.train()
else:
# read from stdin
if args.use_cloudpickle:
import cloudpickle
method_call = cloudpickle.loads(base64.b64decode(args.args_data))
method_call(variant_data)
else:
data = pickle.loads(base64.b64decode(args.args_data))
maybe_iter = concretize(data)
if is_iterable(maybe_iter):
for _ in maybe_iter:
pass
logger.set_snapshot_mode(prev_mode)
logger.set_snapshot_dir(prev_snapshot_dir)
logger.remove_tabular_output(tabular_log_file)
logger.remove_text_output(text_log_file)
logger.pop_prefix()
def run_experiment(
args_data,
variant_data=None,
seed=None,
n_parallel=1,
exp_name=None,
log_dir=None,
snapshot_mode='all',
snapshot_gap=1,
tabular_log_file='progress.csv',
text_log_file='debug.log',
params_log_file='params.json',
variant_log_file='variant.json',
resume_from=None,
plot=False,
log_tabular_only=False,
log_debug_log_only=False,
):
default_log_dir = config.LOG_DIR
now = datetime.datetime.now(dateutil.tz.tzlocal())
# avoid name clashes when running distributed jobs
rand_id = str(uuid.uuid4())[:5]
timestamp = now.strftime('%Y_%m_%d_%H_%M_%S_%f_%Z')
default_exp_name = 'experiment_%s_%s' % (timestamp, rand_id)
if exp_name is None:
exp_name = default_exp_name
if seed is not None:
set_seed(seed)
if n_parallel > 0:
from rllab.sampler import parallel_sampler
parallel_sampler.initialize(n_parallel=n_parallel)
if seed is not None:
parallel_sampler.set_seed(seed)
if plot:
from rllab.plotter import plotter
plotter.init_worker()
if log_dir is None:
log_dir = osp.join(default_log_dir, exp_name)
else:
log_dir = log_dir
tabular_log_file = osp.join(log_dir, tabular_log_file)
text_log_file = osp.join(log_dir, text_log_file)
params_log_file = osp.join(log_dir, params_log_file)
if variant_data is not None:
variant_data = variant_data
variant_log_file = osp.join(log_dir, variant_log_file)
# print(variant_log_file)
# print(variant_data)
logger.log_variant(variant_log_file, variant_data)
else:
variant_data = None
logger.add_text_output(text_log_file)
logger.add_tabular_output(tabular_log_file)
prev_snapshot_dir = logger.get_snapshot_dir()
prev_mode = logger.get_snapshot_mode()
logger.set_snapshot_dir(log_dir)
logger.set_snapshot_mode(snapshot_mode)
logger.set_snapshot_gap(snapshot_gap)
logger.set_log_tabular_only(log_tabular_only)
logger.set_debug_log_only(log_debug_log_only)
logger.push_prefix("[%s] " % exp_name)
if resume_from is not None:
data = joblib.load(resume_from)
assert 'algo' in data
algo = data['algo']
algo.train()
else:
args_data(variant_data)
logger.set_snapshot_mode(prev_mode)
logger.set_snapshot_dir(prev_snapshot_dir)
logger.remove_tabular_output(tabular_log_file)
logger.remove_text_output(text_log_file)
logger.pop_prefix()
