def test_fit_batch_online_pendulum_with_sac():
make_env = lambda: gym.make("Pendulum-v0")
env = AsyncBatchEnv([make_env for _ in range(5)])
eval_env = gym.make("Pendulum-v0")
algo = SAC()
buffer = BatchReplayBuffer(1000, env)
algo.fit_batch_online(
env,
buffer,
n_epochs=1,
n_steps_per_epoch=500,
n_updates_per_epoch=1,
eval_env=eval_env,
logdir="test_data",
tensorboard=False,
)
import gym
from d3rlpy.algos import SAC
from d3rlpy.envs import AsyncBatchEnv
from d3rlpy.online.buffers import BatchReplayBuffer
if __name__ == '__main__':
env = AsyncBatchEnv([lambda: gym.make('Pendulum-v0') for _ in range(10)])
eval_env = gym.make('Pendulum-v0')
# setup algorithm
sac = SAC(batch_size=100, use_gpu=False)
# replay buffer for experience replay
buffer = BatchReplayBuffer(maxlen=100000, env=env)
# start training
sac.fit_batch_online(env,
buffer,
n_epochs=100,
eval_interval=1,
eval_env=eval_env,
n_steps_per_epoch=1000,
n_updates_per_epoch=1000)