"blocks4/task03.pddl",
"blocks4/task04.pddl",
"blocks4/task05.pddl",
"blocks4/task06.pddl",
"blocks4/task07.pddl",
"blocks4/task08.pddl",
"blocks4/task09.pddl",
"blocks4/task10.pddl",
"blocks5/task01.pddl",
"blocks5/task02.pddl",
"blocks5/task03.pddl",
"blocks5/task04.pddl",
"blocks5/task05.pddl",
"blocks5/task06.pddl",
"blocks5/task07.pddl",
"blocks5/task08.pddl",
"blocks5/task09.pddl",
"blocks5/task10.pddl",
],
)
assert len(_CONFIGURATION.problems) == 30
if __name__ == "__main__":
train_wrapper(
args=get_training_args(
configurations=[_CONFIGURATION],
# 10 minutes
max_training_time=10 * 60,
),
domain_name="blocks")
from default_args import get_training_args, DomainAndProblemConfiguration
from train import train_wrapper
_CONFIGURATION = DomainAndProblemConfiguration(
base_directory="../benchmarks/ferry",
domain_pddl="ferry.pddl",
# {2, 3, 4 locations} x {1, 2, 3} cars = 9 problems
problem_pddls=[
"train/ferry-l2-c1.pddl",
"train/ferry-l2-c2.pddl",
"train/ferry-l2-c3.pddl",
"train/ferry-l3-c1.pddl",
"train/ferry-l3-c2.pddl",
"train/ferry-l3-c3.pddl",
"train/ferry-l4-c1.pddl",
"train/ferry-l4-c2.pddl",
"train/ferry-l4-c3.pddl",
],
)
assert len(_CONFIGURATION.problems) == 9
if __name__ == "__main__":
train_wrapper(
args=get_training_args(
configurations=[_CONFIGURATION],
# 3 minutes
max_training_time=3 * 60,
num_folds=5,
),
domain_name="ferry")
def main():
# required for test and train:
# parsed by the main parser - this group is shared
g_parser = argparse.ArgumentParser(add_help=False)
g_parser.add_argument(
'--num-episodes',
type=int,
help=
'number of running episodes (default is 1000 for train, and 5 for test'
)
g_parser.add_argument(
'--build',
default=None,
type=str,
required=True,
help='path of the unity build file, to run inside Unity - enter None')
g_parser.add_argument('--weights-path',
type=str,
required=True,
help='path to weights dir')
g_parser.add_argument('--agent',
choices=['ddpg', 'mddpg', 'maddpg'],
required=True,
help='type of agent')
g_parser.add_argument('--num-agents',
choices=range(1, 9),
default=4,
type=int,
metavar='[1-8]',
help='number of agents (cars)')
g_parser.add_argument('--num-obstacles',
choices=range(0, 17),
default=4,
type=int,
metavar='[0-16]',
help='number of random obstacles')
# general group end
parser = argparse.ArgumentParser(
prog='RL_Multi_agent_Cars',
description='please choose train or test to get specific help'
' (e.g main.py train -h)')
subparsers = parser.add_subparsers(help='two available running modes',
dest='subparser_name')
# define new sub-command
subparsers.add_parser('test', help='run test mode', parents=[g_parser])
# required for train only:
# parse by the train command sub-parser
train_parser = subparsers.add_parser('train',
help='run train mode',
parents=[g_parser])
train_parser.add_argument(
'--save-mem',
action='store_true',
help='save the replay buffer during training for later use',
)
train_parser.add_argument('--scores-avg-window',
choices=range(0, 101),
metavar='[0-100]',
default=50,
type=int,
help='number of last scores to average')
train_parser.add_argument(
'--load-weights',
action='store_true',
help='add this to load weights from previous runs')
train_parser.add_argument(
'--load-mem',
action='store_true',
help='add this to load replay buffer from previous run')
train_parser.add_argument(
'--mem-path',
type=str,
help='path of replay buffer file to load or store')
train_parser.add_argument('--solved-score',
default=40,
type=int,
help='score that complete the episode')
train_parser.add_argument(
'--show-graphics',
action='store_true',
help='add this to show graphics (slows down training)')
train_parser.add_argument(
'--print-agent-loss',
action='store_true',
help='print agent\'s loss after each episode (default=False)')
train_parser.add_argument(
'--save-best-weights',
action='store_true',
help=
'save the best weights so far (by average score). saving directory will be the'
' same as weights-path with suffix \'best\' default=False')
train_parser.add_argument(
'--save-score-log',
action='store_true',
help=
'saves a csv file with the ongoing scores of each episode (default=False)'
)
args = parser.parse_args()
if args.num_episodes is None:
args.num_episodes = 1000 if args.subparser_name == 'train' else 5
env_config = {
'num_agents': args.num_agents,
'num_obstacles': args.num_obstacles,
'setting': 0
}
wrapper_config = vars(args)
wrapper_config['agent'] = select_agent(wrapper_config['agent'])
print('starting {} with arguments:\n{}'.format(args.subparser_name,
wrapper_config))
if args.subparser_name == 'test':
test_wrapper(env_config, wrapper_config)
else:
train_wrapper(env_config, wrapper_config)
"train/zenotravel-cities3-planes3-people3-1826.pddl",
"train/zenotravel-cities3-planes3-people5-4582.pddl",
],
)
assert len(_ZENOTRAVEL_CONFIGURATION.problems) == 10
_GRIPPER_CONFIGURATION = DomainAndProblemConfiguration(
base_directory="../benchmarks/gripper",
domain_pddl="domain.pddl",
# First 3 gripper probs
problem_pddls=[
"problems/gripper-n1.pddl",
"problems/gripper-n2.pddl",
"problems/gripper-n3.pddl",
],
)
assert len(_GRIPPER_CONFIGURATION.problems) == 3
if __name__ == "__main__":
train_wrapper(
args=get_training_args(
configurations=[
_BLOCKSWORLD_CONFIGURATION,
_ZENOTRAVEL_CONFIGURATION,
_GRIPPER_CONFIGURATION,
],
# 15 minutes
max_training_time=15 * 60,
),
domain_name="multi")
"blocks5/task03.pddl",
"blocks5/task04.pddl",
"blocks5/task05.pddl",
],
)
assert len(_BLOCKSWORLD_CONFIGURATION.problems) == 10
_GRIPPER_CONFIGURATION = DomainAndProblemConfiguration(
base_directory="../benchmarks/gripper",
domain_pddl="domain.pddl",
# First 3 gripper probs
problem_pddls=[
"problems/gripper-n1.pddl",
"problems/gripper-n2.pddl",
"problems/gripper-n3.pddl",
],
)
assert len(_GRIPPER_CONFIGURATION.problems) == 3
if __name__ == "__main__":
train_wrapper(
args=get_training_args(
configurations=[
_BLOCKSWORLD_CONFIGURATION,
_GRIPPER_CONFIGURATION,
],
# 10 minutes
max_training_time=10 * 60,
),
domain_name="indepbg")
_ZENOTRAVEL_CONFIGURATION = DomainAndProblemConfiguration(
base_directory="../benchmarks/zenotravel",
domain_pddl="domain.pddl",
# 5 x {2, 3 cities} = 10 Zenotravel problems
problem_pddls=[
"train/zenotravel-cities2-planes1-people3-8798.pddl",
"train/zenotravel-cities2-planes2-people3-9145.pddl",
"train/zenotravel-cities2-planes3-people3-3417.pddl",
"train/zenotravel-cities2-planes4-people2-4892.pddl",
"train/zenotravel-cities2-planes4-people4-6874.pddl",
"train/zenotravel-cities3-planes1-people3-4791.pddl",
"train/zenotravel-cities3-planes2-people3-8752.pddl",
"train/zenotravel-cities3-planes2-people5-7306.pddl",
"train/zenotravel-cities3-planes3-people3-1826.pddl",
"train/zenotravel-cities3-planes3-people5-4582.pddl",
],
)
assert len(_ZENOTRAVEL_CONFIGURATION.problems) == 10
if __name__ == "__main__":
train_wrapper(
args=get_training_args(
configurations=[
_GRIPPER_CONFIGURATION,
_ZENOTRAVEL_CONFIGURATION,
],
# 10 minutes
max_training_time=10 * 60,
),
domain_name="indepgz")
from default_args import get_training_args, DomainAndProblemConfiguration
from train import train_wrapper
_CONFIGURATION = DomainAndProblemConfiguration(
base_directory="../benchmarks/gripper",
domain_pddl="domain.pddl",
# {1, 2, 3 balls} = 3 problems
problem_pddls=[
"problems/gripper-n1.pddl",
"problems/gripper-n2.pddl",
"problems/gripper-n3.pddl",
],
)
assert len(_CONFIGURATION.problems) == 3
if __name__ == "__main__":
train_wrapper(args=get_training_args(
configurations=[_CONFIGURATION],
# 90 seconds
max_training_time=90,
num_bins=3,
))