def generate_experiment_kwargs(variant_spec, command_line_args):
local_dir = command_line_args.local_dir
if command_line_args.mode == 'debug':
local_dir = os.path.join(local_dir, 'debug')
local_dir = os.path.join(local_dir, command_line_args.universe,
command_line_args.domain, command_line_args.task)
resources_per_trial = _normalize_trial_resources(
command_line_args.resources_per_trial, command_line_args.trial_cpus,
command_line_args.trial_gpus, command_line_args.trial_extra_cpus,
command_line_args.trial_extra_gpus)
upload_dir = (os.path.join(
command_line_args.upload_dir, command_line_args.universe,
command_line_args.domain, command_line_args.task)
if command_line_args.upload_dir else None)
datetime_prefix = datetimestamp()
experiment_id = '-'.join((datetime_prefix, command_line_args.exp_name))
variant_spec = add_command_line_args_to_variant_spec(
variant_spec, command_line_args)
if command_line_args.video_save_frequency is not None:
assert 'algorithm_params' in variant_spec
variant_spec['algorithm_params']['config']['video_save_frequency'] = (
command_line_args.video_save_frequency)
def create_trial_name_creator(trial_name_template=None):
if not trial_name_template:
return None
def trial_name_creator(trial):
return trial_name_template.format(trial=trial)
return trial_name_creator
experiment_kwargs = {
'name':
experiment_id,
'resources_per_trial':
resources_per_trial,
'config':
variant_spec,
'local_dir':
local_dir,
'num_samples':
command_line_args.num_samples,
'upload_dir':
upload_dir,
'checkpoint_freq':
(variant_spec['run_params']['checkpoint_frequency']),
'checkpoint_at_end': (variant_spec['run_params']['checkpoint_at_end']),
'max_failures':
command_line_args.max_failures,
'trial_name_creator':
create_trial_name_creator(command_line_args.trial_name_template),
'restore':
command_line_args.restore, # Defaults to None
}
return experiment_kwargs
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 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(
'--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)
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('--seed', type=int, default=1)
parser.add_argument('--run-eagerly',
type=lambda x: bool(strtobool(x)),
help=("Whether to run tensorflow in eager mode."))
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 get_parser(allow_policy_list=False):
parser = argparse.ArgumentParser()
parser.add_argument('--env', type=str, required=True)
parser.add_argument('--randomized', action="store_true", default=False)
parser.add_argument('--use_predictive_model',
action="store_true",
default=False)
parser.add_argument("--observation-mode",
required=True,
type=str,
choices=('pixels', 'pixels_debug'))
parser.add_argument("--reward-type", default='reaching')
parser.add_argument("--single-obj-reward", type=int, required=True)
parser.add_argument("--all-random", type=bool, required=True)
parser.add_argument("--trimodal-positions-choice", type=int, default=0)
parser.add_argument(
"--num-objects",
type=int,
default=3,
)
parser.add_argument("--model-dir", type=str, required=True)
parser.add_argument("--num-execution-per-step", type=int, required=True)
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)
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('--run-eagerly',
type=lambda x: bool(strtobool(x)),
help=("Whether to run tensorflow in eager mode."))
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
