def parse_args():
p = argparse.ArgumentParser()
g = p.add_argument_group('I/O')
g.add_argument('--record_to',
'-rt',
type=str,
default=None,
help="file path to which game will be recorded.")
g.add_argument(
'--playback_from',
'-p',
type=str,
default=None,
help=
'file path from which game will be played-back (overrides record_to)')
g.add_argument('--playback_initial_round',
'-pir',
type=int,
default=0,
help='round in which to start the playback')
g.add_argument('--playback_final_round',
'-pfr',
type=int,
default=1000,
help='round in which to end the playback')
g.add_argument(
'--log_file',
'-l',
type=str,
default=None,
help="a path to which game events are logged. default: game.log")
g.add_argument(
'--output_file',
'-o',
type=str,
default=None,
help=
"a path to a file in which game results are written. default: game.out"
)
g.add_argument('--to_render',
'-r',
type=int,
default=0,
help="whether game should not be rendered")
g.add_argument(
'--render_rate',
'-rr',
type=float,
default=0.1,
help=
'frames per second, note that the policy_wait_time bounds on the rate')
g = p.add_argument_group('Game')
g.add_argument('--board_size',
'-bs',
type=str,
default='(20,60)',
help='a tuple of (height, width)')
g.add_argument('--obstacle_density',
'-od',
type=float,
default=.04,
help='the density of obstacles on the board')
g.add_argument('--policy_wait_time',
'-pwt',
type=float,
default=0.01,
help='seconds to wait for policies to respond with actions')
g.add_argument('--random_food_prob',
'-fp',
type=float,
default=.2,
help='probability of a random food appearing in a round')
g.add_argument(
'--max_item_density',
'-mid',
type=float,
default=.25,
help='maximum item density in the board (not including the players)')
g.add_argument(
'--food_ratio',
'-fr',
type=float,
default=.2,
help=
'the ratio between a corpse and the number of food items it produces')
g.add_argument('--game_duration',
'-D',
type=int,
default=10000,
help='number of rounds in the session')
g.add_argument('--policy_action_time',
'-pat',
type=float,
default=0.01,
help='seconds to wait for agents to respond with actions')
g.add_argument('--policy_learn_time',
'-plt',
type=float,
default=0.1,
help='seconds to wait for agents to improve policy')
g.add_argument(
'--player_init_time',
'-pit',
type=float,
default=PLAYER_INIT_TIME,
help=
'seconds to wait for agents to initialize in the beginning of the session'
)
g = p.add_argument_group('Players')
g.add_argument(
'--score_scope',
'-s',
type=int,
default=1000,
help=
'The score is the average reward during the last score_scope rounds of the session'
)
g.add_argument('--init_player_size',
'-is',
type=int,
default=5,
help='player length at start, minimum is 3')
g.add_argument(
'--policies',
'-P',
type=str,
default=None,
help=
'a string describing the policies to be used in the game, of the form: '
'<policy_name>(<arg=val>,*);+.\n'
'e.g. MyPolicy(layer1=100,layer2=20);YourPolicy(your_params=123)')
args = p.parse_args()
# set defaults
code_path = os.path.split(os.path.abspath(__file__))[0] + os.path.sep
if not args.record_to:
args.__dict__['record_to'] = None
if args.log_file is None:
args.__dict__['log_file'] = code_path + 'game.log'
if args.output_file is None:
args.__dict__['output_file'] = code_path + 'game.out'
if args.playback_from is not None:
args.__dict__['record_to'] = None
args.__dict__['output_file'] = None
args.__dict__['log_file'] = None
args.__dict__['board_size'] = [
int(x) for x in args.board_size[1:-1].split(',')
]
plcs = []
if args.policies is not None: plcs.extend(args.policies.split(';'))
args.__dict__['policies'] = [base_policy.build(p) for p in plcs]
return args
def parse_args():
p = argparse.ArgumentParser()
g = p.add_argument_group('I/O')
g.add_argument('--record_to',
'-rt',
type=str,
default=None,
help="file name to which session will be recorded.")
g.add_argument(
'--playback_from',
'-p',
type=str,
default=None,
help=
'file name from which game will be played-back (overrides record_to)')
g.add_argument(
'--log_file',
'-l',
type=str,
default=None,
help="a path to which game events are logged. default: game.log")
g.add_argument(
'--output_file',
'-o',
type=str,
default=None,
help=
"a path to a file in which game results and model paths are written. default: game.out"
)
g.add_argument(
'--model_folder',
'-mf',
type=str,
default=None,
help=
"a folder to which agents may record their model'. default: this file's folder, .\states\ "
)
g.add_argument(
'--record_folder',
'-rf',
type=str,
default=None,
help=
"a folder to which games are recorded'. default: this file's folder, .\\recordings\ "
)
g.add_argument_group('Playback')
g.add_argument('--initial_playback_round',
'-ipr',
type=int,
default=None,
help='which round to start the playback from')
g.add_argument('--final_playback_round',
'-fpr',
type=int,
default=None,
help='which round to end the playback at')
g.add_argument('--playback_round_skip',
'-prs',
type=int,
default=None,
help='how many rounds to skip over between game playbacks')
g.add_argument(
'--render_rate',
'-rr',
type=float,
default=1.0,
help='seconds to wait between actions when a game is being played back'
)
g = p.add_argument_group('Game')
g.add_argument('--policy_action_time',
'-pat',
type=float,
default=0.01,
help='seconds to wait for agents to respond with actions')
g.add_argument('--policy_learn_time',
'-plt',
type=float,
default=0.1,
help='seconds to wait for agents to improve policy')
g.add_argument('--game_duration',
'-D',
type=int,
default=100,
help='number of rounds')
g.add_argument('--test_or_train',
'-t',
type=str,
default='train',
help='whether this is a training session or a test session')
g.add_argument('--board_initializer',
'-bi',
type=str,
default='RandomBoard',
help='function to be used for board initialization')
g.add_argument('--selfplay',
'-sp',
type=bool,
default=False,
help='whether or not this is a selfplay session')
g.add_argument('--selfplay_time',
'-spt',
type=int,
default=10000,
help='whether or not this is a selfplay session')
g = p.add_argument_group('Agents')
g.add_argument(
'--agents',
'-A',
type=str,
default=None,
help=
'a string describing the two agents to be used in the game, of the form: '
'<agent_name>(<arg=val>,*);<agent_name>(<arg=val>,*)\n'
'e.g. MyAgent(layer1=100,layer2=20);YourAgent(your_params=123)')
args = p.parse_args()
# set defaults for I/O:
code_path = os.path.split(os.path.abspath(__file__))[0] + os.path.sep
if args.log_file is None:
args.__dict__['log_file'] = code_path + 'game.log'
if args.model_folder is None:
args.__dict__['model_folder'] = code_path + 'models' + os.path.sep
if not os.path.exists(args.model_folder):
os.mkdir(args.model_folder)
if args.record_folder is None:
args.__dict__['record_folder'] = code_path + 'recordings' + os.path.sep
if not os.path.exists(args.record_folder):
os.mkdir(args.record_folder)
if args.output_file is None:
args.__dict__['output_file'] = code_path + 'game.out'
if args.playback_from is not None:
args.__dict__['record_to'] = None
args.__dict__['output_file'] = None
args.__dict__['log_file'] = None
args.__dict__['model_folder'] = None
# parse agents:
agents = []
if args.agents is not None: agents.extend(args.agents.split(';'))
if args.playback_from is None:
assert len(agents) == 2, "Wrong Number of Players!"
args.__dict__['agents'] = [base_policy.build(agent) for agent in agents]
args.__dict__['board_initializer'] = base_initializer.find_initializer(
args.__dict__['board_initializer'])
args.__dict__['board_initializer'] = args.__dict__['board_initializer']()
return args