2021-08-31 14:56:53 +02:00
|
|
|
pendulum-sac-fake-gpus:
|
2022-02-04 05:59:56 -08:00
|
|
|
env: Pendulum-v1
|
|
|
|
|
run: SAC
|
|
|
|
|
stop:
|
|
|
|
|
episode_reward_mean: -270
|
|
|
|
|
timesteps_total: 10000
|
|
|
|
|
config:
|
|
|
|
|
# Works for both torch and tf.
|
|
|
|
|
seed: 42
|
|
|
|
|
framework: tf
|
|
|
|
|
horizon: 200
|
|
|
|
|
soft_horizon: false
|
|
|
|
|
Q_model:
|
|
|
|
|
fcnet_activation: relu
|
|
|
|
|
fcnet_hiddens: [ 256, 256 ]
|
|
|
|
|
policy_model:
|
|
|
|
|
fcnet_activation: relu
|
|
|
|
|
fcnet_hiddens: [ 256, 256 ]
|
|
|
|
|
tau: 0.005
|
|
|
|
|
target_entropy: auto
|
|
|
|
|
no_done_at_end: true
|
|
|
|
|
n_step: 1
|
|
|
|
|
rollout_fragment_length: 1
|
|
|
|
|
prioritized_replay: true
|
|
|
|
|
train_batch_size: 256
|
|
|
|
|
target_network_update_freq: 1
|
|
|
|
|
timesteps_per_iteration: 1000
|
|
|
|
|
learning_starts: 256
|
|
|
|
|
num_workers: 0
|
|
|
|
|
metrics_smoothing_episodes: 5
|
2021-08-31 14:56:53 +02:00
|
|
|
|
2022-02-04 05:59:56 -08:00
|
|
|
# 1x batch size (despite 2 GPUs).
|
|
|
|
|
# train_batch_size: 256
|
|
|
|
|
optimization:
|
|
|
|
|
actor_learning_rate: 0.001
|
|
|
|
|
critic_learning_rate: 0.001
|
|
|
|
|
entropy_learning_rate: 0.001
|
2021-08-31 14:56:53 +02:00
|
|
|
|
2022-02-04 05:59:56 -08:00
|
|
|
# Fake 2 GPUs.
|
|
|
|
|
num_gpus: 2
|
|
|
|
|
_fake_gpus: true
|
