2021-02-22 08:30:18 -08:00
|
|
|
hopper_bc:
|
2021-05-04 10:06:19 -07:00
|
|
|
env:
|
|
|
|
|
grid_search:
|
|
|
|
|
- ray.rllib.examples.env.d4rl_env.hopper_random
|
|
|
|
|
#- ray.rllib.examples.env.d4rl_env.hopper_medium
|
|
|
|
|
#- ray.rllib.examples.env.d4rl_env.hopper_expert
|
|
|
|
|
#- ray.rllib.examples.env.d4rl_env.hopper_medium_replay
|
2021-02-22 08:30:18 -08:00
|
|
|
run: CQL
|
|
|
|
|
config:
|
|
|
|
|
# SAC Configs
|
2021-05-04 10:06:19 -07:00
|
|
|
input: d4rl.hopper-random-v0
|
|
|
|
|
#input: d4rl.hopper-medium-v0
|
|
|
|
|
#input: d4rl.hopper-expert-v0
|
|
|
|
|
#input: d4rl.hopper-medium-replay-v0
|
2021-02-22 08:30:18 -08:00
|
|
|
framework: torch
|
2021-05-04 10:06:19 -07:00
|
|
|
soft_horizon: False
|
2021-02-22 08:30:18 -08:00
|
|
|
horizon: 1000
|
|
|
|
|
Q_model:
|
|
|
|
|
fcnet_activation: relu
|
|
|
|
|
fcnet_hiddens: [256, 256, 256]
|
|
|
|
|
policy_model:
|
|
|
|
|
fcnet_activation: relu
|
|
|
|
|
fcnet_hiddens: [256, 256, 256]
|
|
|
|
|
tau: 0.005
|
|
|
|
|
target_entropy: auto
|
|
|
|
|
no_done_at_end: false
|
|
|
|
|
n_step: 1
|
|
|
|
|
rollout_fragment_length: 1
|
2022-05-17 13:43:49 +02:00
|
|
|
replay_buffer_config:
|
|
|
|
|
type: MultiAgentReplayBuffer
|
|
|
|
|
learning_starts: 10
|
2021-02-22 08:30:18 -08:00
|
|
|
train_batch_size: 256
|
2021-05-04 10:06:19 -07:00
|
|
|
target_network_update_freq: 0
|
2022-04-29 22:02:34 +02:00
|
|
|
min_train_timesteps_per_reporting: 1000
|
2021-02-22 08:30:18 -08:00
|
|
|
optimization:
|
2021-05-04 10:06:19 -07:00
|
|
|
actor_learning_rate: 0.0001
|
2021-02-22 08:30:18 -08:00
|
|
|
critic_learning_rate: 0.0003
|
2021-05-04 10:06:19 -07:00
|
|
|
entropy_learning_rate: 0.0001
|
2021-02-22 08:30:18 -08:00
|
|
|
num_workers: 0
|
|
|
|
|
num_gpus: 1
|
|
|
|
|
clip_actions: false
|
|
|
|
|
normalize_actions: true
|
|
|
|
|
evaluation_interval: 1
|
|
|
|
|
metrics_smoothing_episodes: 5
|
|
|
|
|
# CQL Configs
|
|
|
|
|
min_q_weight: 5.0
|
2021-05-04 10:06:19 -07:00
|
|
|
bc_iters: 200000000
|
2021-02-22 08:30:18 -08:00
|
|
|
temperature: 1.0
|
|
|
|
|
num_actions: 10
|
|
|
|
|
lagrangian: False
|
|
|
|
|
evaluation_config:
|
|
|
|
|
input: sampler
|
|
|
|
|
|
