def demo41():
args = Arguments(if_on_policy=True)
args.agent_rl = agent.AgentGaePPO # agent.AgentPPO
import pybullet_envs # for python-bullet-gym
dir(pybullet_envs)
args.env = decorate_env(gym.make('ReacherBulletEnv-v0'))
args.break_step = int(5e4 * 8) # (5e4) 1e5, UsedTime: (400s) 800s
args.repeat_times = 2 ** 3
args.reward_scale = 2 ** 1 # (-15) 18 (30)
args.eval_times1 = 2 ** 2
args.eval_times1 = 2 ** 6
args.rollout_num = 4
train_and_evaluate__multiprocessing(args)
def demo42():
args = Arguments(if_on_policy=True)
args.agent_rl = agent.AgentGaePPO # agent.AgentPPO
import pybullet_envs # for python-bullet-gym
dir(pybullet_envs)
args.env = decorate_env(gym.make('AntBulletEnv-v0'))
args.break_step = int(5e6 * 8) # (1e6) 5e6 UsedTime: 25697s
args.reward_scale = 2 ** -3 #
args.repeat_times = 2 ** 4
args.net_dim = 2 ** 9
args.batch_size = 2 ** 8
args.max_memo = 2 ** 12
args.show_gap = 2 ** 6
args.eval_times1 = 2 ** 2
args.rollout_num = 4
train_and_evaluate__multiprocessing(args)
def demo5():
args = Arguments(if_on_policy=False)
# args.agent_rl = agent.AgentModSAC
args.agent_rl = agent.AgentInterSAC
import pybullet_envs # for python-bullet-gym
dir(pybullet_envs)
args.env = decorate_env(gym.make('AntBulletEnv-v0'))
# args.env = decorate_env(gym.make('ReacherBulletEnv-v0'))
args.break_step = int(1e6 * 8) # (5e5) 1e6, UsedTime: (15,000s) 30,000s
args.reward_scale = 2 ** -2 # (-50) 0 ~ 2500 (3340)
args.max_memo = 2 ** 19
args.net_dim = 2 ** 7 # todo
args.eva_size = 2 ** 5 # for Recorder
args.show_gap = 2 ** 8 # for Recorder
train_and_evaluate(args)
def demo3():
"""DEMO 3: Custom Continuous action env: FinanceStock-v1"""
args = Arguments(if_on_policy=True)
args.agent_rl = agent.AgentGaePPO # PPO+GAE (on-policy)
from eRL.env import FinanceMultiStockEnv
args.env = FinanceMultiStockEnv(if_train=True) # a standard env for ElegantRL, not need decorate_env()
args.env_eval = FinanceMultiStockEnv(if_train=False)
args.break_step = int(5e6) # 5e6 (15e6) UsedTime 3,000s (9,000s)
args.net_dim = 2 ** 8
args.max_step = args.env.max_step
args.max_memo = (args.max_step - 1) * 8
args.batch_size = 2 ** 11
args.repeat_times = 2 ** 4
args.eval_times1 = 2 ** 3
# train_and_evaluate(args)
args.rollout_num = 8
args.if_break_early = False
train_and_evaluate__multiprocessing(args)