def get_parser(allow_policy_list=False):
parser = argparse.ArgumentParser()
parser.add_argument('--universe',
type=str,
choices=UNIVERSES,
default=DEFAULT_UNIVERSE)
parser.add_argument('--domain',
type=str,
choices=AVAILABLE_DOMAINS,
default=DEFAULT_DOMAIN)
parser.add_argument('--task',
type=str,
choices=AVAILABLE_TASKS,
default=DEFAULT_TASK)
parser.add_argument('--sampler',
type=str,
choices=AVAILABLE_SAMPLERS,
default=DEFAULT_SAMPLER)
parser.add_argument('--replay_pool',
type=str,
choices=AVAILABLE_REPLAY_POOLS,
default=DEFAULT_REPLAY_POOL)
parser.add_argument(
'--checkpoint-replay-pool',
type=lambda x: bool(strtobool(x)),
default=None,
help=("Whether a checkpoint should also saved the replay"
" pool. If set, takes precedence over"
" variant['run_params']['checkpoint_replay_pool']."
" Note that the replay pool is saved (and "
" constructed) piece by piece so that each"
" experience is saved only once."))
parser.add_argument('--algorithm',
type=str,
choices=AVAILABLE_ALGORITHMS,
default=DEFAULT_ALGORITHM)
if allow_policy_list:
parser.add_argument('--policy',
type=str,
nargs='+',
choices=('gaussian', ),
default='gaussian')
else:
parser.add_argument('--policy',
type=str,
choices=('gaussian', ),
default='gaussian')
parser.add_argument('--exp-name', type=str, default=datetimestamp())
parser.add_argument('--mode', type=str, default='local')
parser.add_argument('--confirm-remote',
type=lambda x: bool(strtobool(x)),
nargs='?',
const=True,
default=True,
help="Whether or not to query yes/no on remote run.")
parser.add_argument('--video-save-frequency',
type=int,
default=None,
help="Save frequency for videos.")
parser = add_ray_init_args(parser)
parser = add_ray_tune_args(parser)
return parser
def unique_cluster_name(args):
cluster_name_parts = (datetimestamp(''), str(uuid.uuid4())[:6],
args.domain, args.task)
cluster_name = "-".join(cluster_name_parts).lower()
return cluster_name
def launch_experiments_ray(variant_specs,
args,
local_dir,
experiment_fn,
scheduler=None):
import ray
from ray import tune
tune.register_trainable('mujoco-runner', experiment_fn)
resources_per_trial = _normalize_trial_resources(args.resources_per_trial,
args.trial_cpus,
args.trial_gpus,
args.trial_extra_cpus,
args.trial_extra_gpus)
if 'local' in args.mode or 'debug' in args.mode:
resources = args.resources or {}
if 'debug' in args.mode:
# Require a debug resource for each trial, so that we never run
# more than one trial at a time. This makes debugging easier, since
# the debugger stdout behaves more reasonably with single process.
# TODO(hartikainen): Change this from 'extra_gpu' to
# 'debug-resource' once tune supports custom resources.
# See: https://github.com/ray-project/ray/pull/2979.
resources['extra_gpu'] = 1
resources_per_trial['extra_gpu'] = 1
ray.init(resources=resources, num_cpus=args.cpus, num_gpus=args.gpus)
else:
ray.init(redis_address=ray.services.get_node_ip_address() + ':6379')
datetime_prefix = datetimestamp()
experiment_id = '-'.join((datetime_prefix, args.exp_name))
tune.run_experiments(
{
"{}-{}".format(experiment_id, i): {
'run':
'mujoco-runner',
'resources_per_trial':
resources_per_trial,
'config':
variant_spec,
'local_dir':
local_dir,
'num_samples':
args.num_samples,
'upload_dir':
args.upload_dir,
'checkpoint_freq':
(args.checkpoint_frequency if args.checkpoint_frequency
is not None else variant_spec['run_params'].get(
'checkpoint_frequency', 0)),
'checkpoint_at_end':
(args.checkpoint_at_end if args.checkpoint_at_end
is not None else variant_spec['run_params'].get(
'checkpoint_at_end', True)),
'restore':
args.restore, # Defaults to None
}
for i, variant_spec in enumerate(variant_specs)
},
scheduler=scheduler,
)
def get_parser(allow_policy_list=False):
parser = argparse.ArgumentParser()
parser.add_argument('--universe',
type=str,
choices=UNIVERSES,
default=None)
parser.add_argument('--domain',
type=str,
choices=AVAILABLE_DOMAINS,
default=None)
parser.add_argument('--task',
type=str,
choices=AVAILABLE_TASKS,
default=DEFAULT_TASK)
parser.add_argument('--num-samples', type=int, default=1)
parser.add_argument('--resources',
type=json.loads,
default=None,
help=("Resources to allocate to ray process. Passed"
" to `ray.init`."))
parser.add_argument('--cpus',
type=int,
default=None,
help=("Cpus to allocate to ray process. Passed"
" to `ray.init`."))
parser.add_argument('--gpus',
type=int,
default=None,
help=("Gpus to allocate to ray process. Passed"
" to `ray.init`."))
parser.add_argument('--resources-per-trial',
type=json.loads,
default={},
help=("Resources to allocate for each trial. Passed"
" to `tune.run_experiments`."))
parser.add_argument('--trial-cpus',
type=int,
default=multiprocessing.cpu_count(),
help=("Resources to allocate for each trial. Passed"
" to `tune.run_experiments`."))
parser.add_argument('--trial-gpus',
type=float,
default=None,
help=("Resources to allocate for each trial. Passed"
" to `tune.run_experiments`."))
parser.add_argument(
'--trial-extra-cpus',
type=int,
default=None,
help=("Extra CPUs to reserve in case the trials need to"
" launch additional Ray actors that use CPUs."))
parser.add_argument(
'--trial-extra-gpus',
type=float,
default=None,
help=("Extra GPUs to reserve in case the trials need to"
" launch additional Ray actors that use GPUs."))
parser.add_argument(
'--checkpoint-frequency',
type=int,
default=None,
help=("Save the training checkpoint every this many"
" epochs. If set, takes precedence over"
" variant['run_params']['checkpoint_frequency']."))
parser.add_argument('--checkpoint-at-end',
type=lambda x: bool(strtobool(x)),
default=None,
help=("Whether a checkpoint should be saved at the end"
" of training. If set, takes precedence over"
" variant['run_params']['checkpoint_at_end']."))
parser.add_argument(
'--checkpoint-replay-pool',
type=lambda x: bool(strtobool(x)),
default=None,
help=("Whether a checkpoint should also saved the replay"
" pool. If set, takes precedence over"
" variant['run_params']['checkpoint_replay_pool']."
" Note that the replay pool is saved (and "
" constructed) piece by piece so that each"
" experience is saved only once."))
parser.add_argument('--restore',
type=str,
default=None,
help=("Path to checkpoint. Only makes sense to set if"
" running 1 trial. Defaults to None."))
if allow_policy_list:
parser.add_argument('--policy',
type=str,
nargs='+',
choices=('gaussian', ),
default='gaussian')
else:
parser.add_argument('--policy',
type=str,
choices=('gaussian', ),
default='gaussian')
parser.add_argument('--env', type=str, default='gym-swimmer-default')
parser.add_argument('--exp-name', type=str, default=datetimestamp())
parser.add_argument('--mode', type=str, default='local')
parser.add_argument('--log-dir', type=str, default=None)
parser.add_argument('--upload-dir',
type=str,
default='',
help=("Optional URI to sync training results to (e.g."
" s3://<bucket> or gs://<bucket>)."))
parser.add_argument("--confirm-remote",
type=lambda x: bool(strtobool(x)),
nargs='?',
const=True,
default=True,
help="Whether or not to query yes/no on remote run.")
return parser
def get_parser(allow_policy_list=False):
parser = argparse.ArgumentParser()
parser.add_argument('--universe',
type=str,
choices=AVAILABLE_UNIVERSES,
default=DEFAULT_UNIVERSE)
parser.add_argument('--domain', type=str, default=DEFAULT_DOMAIN)
parser.add_argument('--task', type=str, default=DEFAULT_TASK)
parser.add_argument('--task-evaluation', type=str, default=DEFAULT_TASK)
parser.add_argument('--n-epochs', type=int, default=1000)
parser.add_argument(
'--checkpoint-replay-pool',
type=lambda x: bool(strtobool(x)),
default=None,
help=("Whether a checkpoint should also saved the replay"
" pool. If set, takes precedence over"
" variant['run_params']['checkpoint_replay_pool']."
" Note that the replay pool is saved (and "
" constructed) piece by piece so that each"
" experience is saved only once."))
parser.add_argument('--algorithm',
type=str,
choices=AVAILABLE_ALGORITHMS,
default=DEFAULT_ALGORITHM)
if allow_policy_list:
parser.add_argument('--policy',
type=str,
nargs='+',
choices=('gaussian', ),
default='gaussian')
else:
parser.add_argument('--policy',
type=str,
choices=('gaussian', ),
default='gaussian')
parser.add_argument('--exp-name', type=str, default=datetimestamp())
parser.add_argument('--mode', type=str, default='local')
parser.add_argument('--confirm-remote',
type=lambda x: bool(strtobool(x)),
nargs='?',
const=True,
default=True,
help="Whether or not to query yes/no on remote run.")
parser.add_argument('--video-save-frequency',
type=int,
default=None,
help="Save frequency for videos.")
parser.add_argument('--save-training-video-frequency',
type=int,
default=None,
help="Save frequency for training videos.")
from softlearning.preprocessors.utils import PREPROCESSOR_FUNCTIONS
parser.add_argument('--preprocessor-type',
type=str,
default=None,
help="Preprocessor type for observations.",
choices=list(PREPROCESSOR_FUNCTIONS.keys()))
parser.add_argument('--vision',
type=lambda x: bool(strtobool(x)),
default=True)
parser = add_ray_init_args(parser)
parser = add_ray_tune_args(parser)
return parser