def __init__(
self,
name,
chief_handle,
game_cls,
agent_bet_set,
algo_name,
starting_stack_sizes=None,
):
"""
Args:
name (str): Under this name all logs, data, and checkpoints will appear.
chief_handle (ChiefBase): Reference to chief worker
game_cls (PokerEnv subclass): Class (not instance) to be trained in.
agent_bet_set (iterable): Choosing a bet-set from bet_sets.py is recommended. If solving a
Limit poker game, this value will not be considered, but must still
be passed. Just set this to any list of floats (e.g. [0.0])
starting_stack_sizes (list of ints): For each stack size in this list, a CFR strategy will be computed.
Results are logged individually and averaged (uniform).
If None, takes the default for the game.
"""
self._name = name
self._n_seats = 2
self._chief_handle = chief_handle
if starting_stack_sizes is None:
self._starting_stack_sizes = [game_cls.DEFAULT_STACK_SIZE]
else:
self._starting_stack_sizes = copy.deepcopy(starting_stack_sizes)
self._game_cls_str = game_cls.__name__
self._env_args = [
game_cls.ARGS_CLS(
n_seats=self._n_seats,
starting_stack_sizes_list=[
start_chips for _ in range(self._n_seats)
],
bet_sizes_list_as_frac_of_pot=agent_bet_set,
) for start_chips in self._starting_stack_sizes
]
self._env_bldrs = [
HistoryEnvBuilder(env_cls=get_env_cls_from_str(self._game_cls_str),
env_args=self._env_args[s])
for s in range(len(self._starting_stack_sizes))
]
self._trees = [
PublicTree(
env_bldr=self._env_bldrs[idx],
stack_size=self._env_args[idx].starting_stack_sizes_list,
stop_at_street=None) for idx in range(len(self._env_bldrs))
]
for tree in self._trees:
tree.build_tree()
print("Tree with stack size", tree.stack_size, "has", tree.n_nodes,
"nodes out of which", tree.n_nonterm, "are non-terminal.")
self._algo_name = algo_name
self._exps_curr_total = [
self._chief_handle.create_experiment(
self._name + "_Curr_S" + str(self._starting_stack_sizes[s]) +
"_total_" + self._algo_name)
for s in range(len(self._starting_stack_sizes))
]
self._exps_avg_total = [
self._chief_handle.create_experiment(
self._name + "_Avg_total_S" +
str(self._starting_stack_sizes[s]) + "_" + self._algo_name)
for s in range(len(self._starting_stack_sizes))
]
self._exp_all_averaged_curr_total = self._chief_handle.create_experiment(
self._name + "_Curr_total_averaged_" + self._algo_name)
self._exp_all_averaged_avg_total = self._chief_handle.create_experiment(
self._name + "_Avg_total_averaged_" + self._algo_name)
self._iter_counter = None
def get_env_builder_rlbr(t_prof):
env_bldr_cls = get_builder_from_str(t_prof.env_builder_cls_str)
return env_bldr_cls(env_cls=get_env_cls_from_str(t_prof.game_cls_str),
env_args=t_prof.module_args["rlbr"].get_rlbr_env_args(
agents_env_args=t_prof.module_args["env"]))