from magent.builtin.tf_model import DeepQNetwork
models.append(
DeepQNetwork(env,
tiger_handle,
args.name,
batch_size=batch_size,
memory_size=2**20,
learning_rate=4e-4))
step_batch_size = None
elif args.alg == 'drqn':
from magent.builtin.tf_model import DeepRecurrentQNetwork
models.append(
DeepRecurrentQNetwork(env,
tiger_handle,
"tiger",
batch_size=batch_size / unroll,
unroll_step=unroll,
memory_size=20000,
learning_rate=4e-4))
step_batch_size = None
elif args.alg == 'a2c':
from magent.builtin.mx_model import AdvantageActorCritic
step_batch_size = int(10 * args.map_size * args.map_size * 0.01)
models.append(
AdvantageActorCritic(env,
tiger_handle,
"tiger",
batch_size=step_batch_size,
learning_rate=1e-2))
else:
raise NotImplementedError
handles[0],
"battle",
batch_size=batch_size,
learning_rate=3e-4,
memory_size=2**21,
target_update=target_update,
train_freq=train_freq,
eval_obs=eval_obs))
elif args.alg == 'drqn':
from magent.builtin.tf_model import DeepRecurrentQNetwork
models.append(
DeepRecurrentQNetwork(env,
handles[0],
"battle",
learning_rate=3e-4,
batch_size=batch_size / unroll_step,
unroll_step=unroll_step,
memory_size=2 * 8 * 625,
target_update=target_update,
train_freq=train_freq,
eval_obs=eval_obs))
else:
# see train_against.py to know how to use a2c
raise NotImplementedError
models.append(models[0])
# load if
savedir = 'save_model'
if args.load_from is not None:
start_from = args.load_from
print("load ... %d" % start_from)
models.append(
DeepQNetwork(env,
handles[0],
"selfplay",
batch_size=batch_size,
memory_size=2**20,
target_update=target_update,
train_freq=train_freq,
eval_obs=eval_obs))
elif args.alg == 'drqn':
models.append(
DeepRecurrentQNetwork(env,
handles[0],
"selfplay",
batch_size=batch_size / unroll_step,
unroll_step=unroll_step,
memory_size=2 * 8 * 625,
target_update=target_update,
train_freq=train_freq,
eval_obs=eval_obs))
else:
raise NotImplementedError
models.append(models[0])
# load if
savedir = 'save_model'
if args.load_from is not None:
start_from = args.load_from
print("load ... %d" % start_from)
for model in models: