def __init__(self):
self.context_params = ContextParams()
self.args = ArgsProvider(define_params=[
("handicap_level", 0), ("latest_start", 1000),
("latest_start_decay", 0.7), ("fs_ai", 50), ("fs_opponent", 50),
("ai_type",
dict(type=str,
choices=[
"AI_SIMPLE", "AI_HIT_AND_RUN", "AI_NN", "AI_FLAG_NN",
"AI_TD_NN"
],
default="AI_NN")),
("opponent_type",
dict(type=str,
choices=[
"AI_SIMPLE", "AI_HIT_AND_RUN", "AI_FLAG_SIMPLE",
"AI_TD_BUILT_IN"
],
default="AI_SIMPLE")),
("max_tick", dict(type=int, default=30000, help="Maximal tick")),
("mcts_threads", 64), ("seed", 0), ("simple_ratio", -1),
("ratio_change", 0), ("actor_only", dict(action="store_true"))
],
more_params=["batchsize", "T"],
child_providers=[self.context_params.args])
def __init__(self):
self.args = ArgsProvider(call_from=self,
define_params=[
("num_minibatch", 5000),
("num_episode", 10000),
("tqdm", dict(action="store_true")),
])
def __init__(self):
self.args = ArgsProvider(
call_from = self,
define_params = [
("stats", dict(type=str, choices=["rewards", "winrate"], default="rewards")),
("num_eval", 500)
]
)
def __init__(self):
super(EvaluationProcess, self).__init__()
self.server = ParameterServer(2)
self.args = ArgsProvider(call_from=self,
define_params=[
("eval_freq", 10),
("eval_gpu", 1),
])
self.count = 0
def __init__(self):
self.args = ArgsProvider(
call_from=self,
define_args=[("num_games", 1024), ("batchsize", 128),
("game_multi", dict(type=int, default=None)),
("T", 6), ("eval", dict(action="store_true")),
("wait_per_group", dict(action="store_true")),
("verbose_comm", dict(action="store_true")),
("verbose_collector", dict(action="store_true"))],
on_get_args=self._on_get_args)
def __init__(self, model_class):
self.model_class = model_class
self.args = ArgsProvider(call_from=self,
define_params=[("gpu",
dict(type=str,
help="gpu to use",
default=None)),
("load",
dict(type=str,
help="load model",
default=None))])
def __init__(self):
self.context_args = ContextArgs()
self.args = ArgsProvider(call_from=self,
define_args=[
("frame_skip", 4), ("hist_len", 4),
("rom_file", "pong.bin"),
("actor_only", dict(action="store_true")),
("reward_clip", 1),
("rom_dir", os.path.dirname(__file__))
],
more_args=["batchsize", "T"],
child_providers=[self.context_args.args])
def __init__(self):
self.timer = RLTimer()
self.last_time = None
self.args = ArgsProvider(
call_from=self,
define_params=[
("freq_update", 1),
("record_dir", "./record"),
("save_prefix", "save"),
("save_dir", dict(type=str, default=None)),
],
more_params=["num_games", "batchsize", "num_minibatch"],
on_get_params=self._on_get_params)
def __init__(self, mi=None, args=None):
if args is None:
self.args = ArgsProvider(call_from=self,
define_params=self._params(),
on_get_params=self._init)
else:
self.args = args
self._init(args)
# Accumulated errors.
self.stats = defaultdict(lambda: Stats())
self._cb = {}
if mi is not None:
self.model_interface = mi
def __init__(self):
self.args = ArgsProvider(
call_from=self,
define_params=[
("sample_policy",
dict(type=str,
choices=["epsilon-greedy", "multinomial", "uniform"],
help="Sample policy",
default="epsilon-greedy")),
("sample_node", dict(type=str, default="pi")),
("greedy", dict(action="store_true")),
("epsilon",
dict(type=float,
help="Used in epsilon-greedy approach",
default=0.00))
])