|
from easydict import EasyDict |
|
|
|
lunarlander_ppo_config = dict( |
|
exp_name='lunarlander_gcl_seed0', |
|
env=dict( |
|
collector_env_num=8, |
|
evaluator_env_num=8, |
|
env_id='LunarLander-v2', |
|
n_evaluator_episode=8, |
|
stop_value=200, |
|
), |
|
reward_model=dict( |
|
learning_rate=0.001, |
|
input_size=9, |
|
batch_size=32, |
|
continuous=False, |
|
update_per_collect=20, |
|
), |
|
policy=dict( |
|
cuda=False, |
|
action_space='discrete', |
|
recompute_adv=True, |
|
model=dict( |
|
obs_shape=8, |
|
action_shape=4, |
|
action_space='discrete', |
|
), |
|
learn=dict( |
|
update_per_collect=8, |
|
batch_size=800, |
|
learning_rate=0.001, |
|
value_weight=0.5, |
|
entropy_weight=0.01, |
|
clip_ratio=0.2, |
|
adv_norm=True, |
|
), |
|
collect=dict( |
|
|
|
|
|
|
|
model_path='model_path_placeholder', |
|
|
|
|
|
|
|
collector_logit=True, |
|
n_sample=800, |
|
unroll_len=1, |
|
discount_factor=0.99, |
|
gae_lambda=0.95, |
|
), |
|
), |
|
) |
|
lunarlander_ppo_config = EasyDict(lunarlander_ppo_config) |
|
main_config = lunarlander_ppo_config |
|
lunarlander_ppo_create_config = dict( |
|
env=dict( |
|
type='lunarlander', |
|
import_names=['dizoo.box2d.lunarlander.envs.lunarlander_env'], |
|
), |
|
env_manager=dict(type='subprocess'), |
|
policy=dict(type='ppo'), |
|
reward_model=dict(type='guided_cost'), |
|
) |
|
lunarlander_ppo_create_config = EasyDict(lunarlander_ppo_create_config) |
|
create_config = lunarlander_ppo_create_config |
|
|
|
if __name__ == "__main__": |
|
from ding.entry import serial_pipeline_guided_cost |
|
serial_pipeline_guided_cost([main_config, create_config], seed=0) |
|
|