from ray import tune
from toolbox.dice.dice import DiCETrainer
from toolbox.marl import get_marl_env_config
from toolbox.train import train, get_train_parser
if __name__ == '__main__':
parser = get_train_parser()
parser.add_argument("--num-agents", type=int, default=10)
args = parser.parse_args()
env_name = args.env_name
exp_name = "{}-{}".format(args.exp_name, env_name)
stop = int(5e7)
config = {
"num_sgd_iter": 10,
"num_envs_per_worker": 1,
"entropy_coeff": 0.001,
"lambda": 0.95,
"lr": 2.5e-4,
# 'sample_batch_size': 200 if large else 50,
# 'sgd_minibatch_size': 100 if large else 64,
# 'train_batch_size': 10000 if large else 2048,
"num_gpus": 1,
"num_cpus_per_worker": 1,
"num_cpus_for_driver": 2,
'num_workers': 16
}
from ray import tune
from toolbox.train import train, get_train_parser
if __name__ == '__main__':
args = get_train_parser().parse_args()
exp_name = args.exp_name # It's "12230-ppo..." previously....
env_name = args.env_name
stop = int(1e6)
sac_config = {
"env":
tune.grid_search([
"HalfCheetah-v3",
# "Walker2d-v3",
"Ant-v3",
# "Hopper-v3",
"Humanoid-v3"
]),
"horizon":
1000,
"sample_batch_size":
1,
"train_batch_size":
256,
"target_network_update_freq":
1,
"timesteps_per_iteration":
1000,
"learning_starts":
