File size: 3,005 Bytes
079c32c |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 |
from easydict import EasyDict
from ding.entry import serial_pipeline_dream
# environment hypo
env_id = 'Hopper-v2'
obs_shape = 11
action_shape = 3
# gpu
cuda = True
main_config = dict(
exp_name='hopper_stevesac_mbpo_seed0',
env=dict(
env_id=env_id,
norm_obs=dict(use_norm=False, ),
norm_reward=dict(use_norm=False, ),
collector_env_num=1,
evaluator_env_num=8,
n_evaluator_episode=8,
stop_value=100000,
),
policy=dict(
cuda=cuda,
# it is better to put random_collect_size in policy.other
random_collect_size=10000,
model=dict(
obs_shape=obs_shape,
action_shape=action_shape,
twin_critic=True,
action_space='reparameterization',
actor_head_hidden_size=256,
critic_head_hidden_size=256,
),
learn=dict(
ensemble_size=7,
update_per_collect=20,
batch_size=256,
learning_rate_q=3e-4,
learning_rate_policy=3e-4,
learning_rate_alpha=3e-4,
ignore_done=False,
target_theta=0.005,
discount_factor=0.99,
alpha=0.2,
reparameterization=True,
auto_alpha=False,
),
collect=dict(
n_sample=1,
unroll_len=1,
),
command=dict(),
eval=dict(evaluator=dict(eval_freq=500, )), # w.r.t envstep
other=dict(
# environment buffer
replay_buffer=dict(replay_buffer_size=1000000, periodic_thruput_seconds=60),
),
),
world_model=dict(
eval_freq=250, # w.r.t envstep
train_freq=250, # w.r.t envstep
cuda=cuda,
rollout_length_scheduler=dict(
type='linear',
rollout_start_step=20000,
rollout_end_step=40000,
rollout_length_min=1,
rollout_length_max=3,
),
model=dict(
ensemble_size=7,
elite_size=5,
state_size=obs_shape, # has to be specified
action_size=action_shape, # has to be specified
reward_size=1,
hidden_size=200,
use_decay=True,
batch_size=256,
holdout_ratio=0.1,
max_epochs_since_update=5,
deterministic_rollout=True,
),
),
)
main_config = EasyDict(main_config)
create_config = dict(
env=dict(
type='mbmujoco',
import_names=['dizoo.mujoco.envs.mujoco_env'],
),
env_manager=dict(type='subprocess'),
policy=dict(
type='stevesac',
import_names=['ding.policy.mbpolicy.mbsac'],
),
replay_buffer=dict(type='naive', ),
world_model=dict(
type='mbpo',
import_names=['ding.world_model.mbpo'],
),
)
create_config = EasyDict(create_config)
if __name__ == '__main__':
serial_pipeline_dream((main_config, create_config), seed=0, max_env_step=100000)
|