def get_agent(env,
nsteps=5,
nstack=1,
total_timesteps=int(80e6),
vf_coef=0.5,
ent_coef=0.01,
max_grad_norm=0.5,
lr=7e-4,
epsilon=1e-5,
alpha=0.99):
# Note: nstack=1 since frame_stack=True, during training frame_stack=False
agent = Agent(Network=CNN,
ob_space=env.observation_space,
ac_space=env.action_space,
nenvs=1,
nsteps=nsteps,
nstack=nstack,
ent_coef=ent_coef,
vf_coef=vf_coef,
max_grad_norm=max_grad_norm,
lr=lr,
alpha=alpha,
epsilon=epsilon,
total_timesteps=total_timesteps)
return agent
import gym
import time
from a2c import Agent
import sys
if len(sys.argv) != 2:
raise ValueError('test or train?')
env = gym.make('CartPole-v1')
agent = Agent(env)
if sys.argv[1] == 'train':
agent.train()
else:
agent.test()