def __init__(self, t_prof, chief_handle, eval_agent_cls):
super().__init__(t_prof=t_prof)
self._args = t_prof.module_args["rlbr"]
self._env_bldr = rl_util.get_env_builder(t_prof=t_prof)
self._chief_handle = chief_handle
self._eval_agent_cls = eval_agent_cls
self._eval_env_bldr = _util.get_env_builder_rlbr(t_prof=t_prof)
self._ddqns = [None for _ in range(self._eval_env_bldr.N_SEATS)]
self._rlbr_seat_id = None
self._agent_seat_id = None
self._rlbr_env_wrapper = None
self._opponent = None
self._buf = None
self._br_memory_saver = None
if t_prof.nn_type == "recurrent":
from PokerRL.rl.buffers.CircularBufferRNN import CircularBufferRNN
from PokerRL.rl.buffers.BRMemorySaverRNN import BRMemorySaverRNN
self.CircularBufferCls = CircularBufferRNN
self.BRMemorySaverCls = BRMemorySaverRNN
elif t_prof.nn_type == "feedforward":
from PokerRL.rl.buffers.CircularBufferFLAT import CircularBufferFLAT
from PokerRL.rl.buffers.BRMemorySaverFLAT import BRMemorySaverFLAT
self.CircularBufferCls = CircularBufferFLAT
self.BRMemorySaverCls = BRMemorySaverFLAT
else:
raise ValueError(t_prof.nn_type)
def __init__(self, t_prof, chief_handle, eval_agent_cls):
super().__init__(
t_prof=t_prof,
eval_env_bldr=_util.get_env_builder_rlbr(t_prof=t_prof),
chief_handle=chief_handle,
evaluator_name="RL-BR",
log_conf_interval=True)
assert self._eval_env_bldr.N_SEATS == 2, "only works for 2 players at the moment"
self._args = t_prof.module_args["rlbr"]
self._eval_agent = eval_agent_cls(t_prof=t_prof)
self._la_handles_0 = None
self._la_handles_1 = None
self._ps_handle = None
if self._t_prof.log_verbose and self._args.n_brs_to_train > 1:
self._exps_br_quality = {
eval_mode: [[
self._ray.get(
self._ray.remote(
self._chief_handle.create_experiment,
self._t_prof.name + " " + eval_mode + "_stack_" +
str(stack_size[0]) + ": " + self._evaluator_name +
" RLBR Quality" + "_brID" + str(_br_id)))
for _br_id in range(self._args.n_brs_to_train)
] for stack_size in self._t_prof.eval_stack_sizes]
for eval_mode in self._t_prof.eval_modes_of_algo
}
def __init__(self, t_prof, chief_handle):
super().__init__(t_prof=t_prof, chief_handle=chief_handle)
self._args = t_prof.module_args["rlbr"]
self._eval_env_bldr = _util.get_env_builder_rlbr(t_prof=t_prof)
self._env_bldr = self._eval_env_bldr # Override base class variable
if self._t_prof.log_verbose:
self._exp_mem_usage = self._ray.get(
self._ray.remote(self._chief_handle.create_experiment,
self._t_prof.name + "_RL-BR_PS_Memory_Usage"))
self._nets = [None for _ in range(self._env_bldr.N_SEATS)]
self._eps = [None for _ in range(self._env_bldr.N_SEATS)]
self._optims = [None for _ in range(self._env_bldr.N_SEATS)]
def __init__(self, t_prof, chief_handle, eval_agent_cls):
super().__init__(
t_prof=t_prof,
eval_env_bldr=_util.get_env_builder_rlbr(t_prof=t_prof),
chief_handle=chief_handle,
eval_type="RL-BR",
log_conf_interval=True)
assert self._eval_env_bldr.N_SEATS == 2, "only works for 2 players at the moment"
self._args = t_prof.module_args["rlbr"]
self._eval_agent = eval_agent_cls(t_prof=t_prof)
self._la_handles = None
self._ps_handle = None