import time
import agnes
env_name = "InvertedDoublePendulum-v2" # "Swimmer-v2" # "BreakoutNoFrameskip-v4" #
if __name__ == '__main__':
envs = agnes.make_vec_env(env_name)
runner = agnes.Single(envs, agnes.PPORND, agnes.MLP)
runner.log(agnes.TensorboardLogger('.logs/'), agnes.log)
runner.run()
env = agnes.make_env(env_name)
agnes.common.Visualize(runner.worker, env).run()
import agnes
def test_config():
return dict(timesteps=30000,
nsteps=128,
nminibatches=4,
gamma=1.0,
lam=0.95,
noptepochs=4,
max_grad_norm=2.0,
learning_rate=lambda x: 2.5e-4 * x,
cliprange=lambda x: 0.1 * x,
vf_coef=0.5,
ent_coef=.01)
if __name__ == '__main__':
env = agnes.make_vec_env('CartPole-v0', envs_num=2)
runner = agnes.DistributedMPI(env,
agnes.PPO,
agnes.MLP,
config=test_config())
runner.log(agnes.log)
runner.run()
import agnes
import torch
import time
env_name3 = "Walker2d-v2"
if __name__ == '__main__':
envs = agnes.make_vec_env(env_name3, envs_num=8)
runner = agnes.Distributed(envs, agnes.PPO, agnes.MLP)
runner.log(agnes.log, agnes.TensorboardLogger())
runner.run()
del runner
def test_vec():
envs = agnes.make_vec_env('Pendulum-v0')
runner = agnes.Single(envs, agnes.PPO, agnes.MLP, config=test_config())
runner.log(agnes.log)
runner.run()
import agnes
import time
env_name = "BreakoutNoFrameskip-v4"
if __name__ == '__main__':
envs = agnes.make_vec_env(env_name, envs_num=4)
runner = agnes.Single(envs, agnes.PPO, agnes.CNN)
runner.log(agnes.TensorboardLogger(), agnes.log)
runner.run()
import time
import agnes
env_name = "BreakoutNoFrameskip-v4"
if __name__ == '__main__':
env = agnes.make_vec_env(env_name,
envs_num=4,
config={"frame_stack": True})
runner = agnes.Single(env, agnes.PPO, agnes.LSTMCNN)
# runner.worker.load("examples/distributed_rnn/Breakout.pth")
# runner.trainer.load("examples/distributed_rnn/Breakout.pth")
runner.log(agnes.TensorboardLogger(".logs/"), agnes.CsvLogger(".logs/"))
runner.save_every("temp.pth", int(1e6))
runner.run()
runner.save("final.pth")