def MCTS_agent_fn(arena_id, env):
args_mcts = dict(recursive=False,
max_episode_length=5,
num_simulation=100,
max_rollout_steps=5,
c_init=0.1,
c_base=1000000,
num_samples=1,
num_processes=1,
logging=True,
logging_graphs=True)
args_mcts['agent_id'] = 1
args_mcts['char_index'] = 0
return MCTS_agent(**args_mcts)
max_episode_length=5,
num_simulation=100,
max_rollout_steps=5,
c_init=0.1,
c_base=1000000,
num_samples=1,
num_processes=1,
logging=True,
logging_graphs=True)
args_agent1 = {'agent_id': 1, 'char_index': 0}
args_agent2 = {'agent_id': 2, 'char_index': 1}
args_agent1.update(args_common)
args_agent2.update(args_common)
args_agent2.update({'recursive': False})
agents = [lambda x, y: MCTS_agent(**args_agent1), lambda x, y: MCTS_agent(**args_agent2)]
arena = ArenaMP(args.max_episode_length, id_run, env_fn, agents)
for iter_id in range(num_tries):
steps_list, failed_tasks = [], []
if not os.path.isfile(args.record_dir + '/results_{}.pik'.format(0)):
test_results = {}
else:
test_results = pickle.load(open(args.record_dir + '/results_{}.pik'.format(0), 'rb'))
current_tried = iter_id
for episode_id in episode_ids:
curr_log_file_name = args.record_dir + '/logs_agent_{}_{}_{}.pik'.format(
env_task_set[episode_id]['task_id'],
num_simulation=100,
max_rollout_steps=5,
c_init=0.1,
c_base=1000000,
num_samples=1,
num_processes=1,
logging=True,
logging_graphs=True)
args_agent1 = {'agent_id': 1, 'char_index': 0}
args_agent2 = {'agent_id': 2, 'char_index': 1}
args_agent1.update(args_common)
args_agent2.update(args_common)
args_agent2.update({'recursive': True})
agents = [
lambda x, y: MCTS_agent(**args_agent1),
lambda x, y: Random_agent(**args_agent2)
]
arena = ArenaMP(args.max_episode_length, id_run, env_fn, agents)
for iter_id in range(num_tries):
steps_list, failed_tasks = [], []
if not os.path.isfile(args.record_dir + '/results_{}.pik'.format(0)):
test_results = {}
else:
test_results = pickle.load(
open(args.record_dir + '/results_{}.pik'.format(0), 'rb'))
current_tried = iter_id
for episode_id in episode_ids: