2019-07-29 15:02:32 -07:00
|
|
|
# This can reach 9k reward in 2 hours on a Titan XP GPU
|
|
|
|
|
# with 16 workers and 8 envs per worker.
|
|
|
|
|
halfcheetah-appo:
|
|
|
|
|
env: HalfCheetah-v2
|
|
|
|
|
run: APPO
|
|
|
|
|
stop:
|
|
|
|
|
time_total_s: 10800
|
|
|
|
|
config:
|
|
|
|
|
vtrace: True
|
|
|
|
|
gamma: 0.99
|
|
|
|
|
lambda: 0.95
|
2020-03-14 12:05:04 -07:00
|
|
|
rollout_fragment_length: 512
|
2019-07-29 15:02:32 -07:00
|
|
|
train_batch_size: 4096
|
|
|
|
|
num_workers: 16
|
|
|
|
|
num_gpus: 1
|
|
|
|
|
broadcast_interval: 1
|
|
|
|
|
max_sample_requests_in_flight_per_worker: 1
|
|
|
|
|
num_data_loader_buffers: 1
|
|
|
|
|
num_envs_per_worker: 32
|
|
|
|
|
minibatch_buffer_size: 16
|
|
|
|
|
num_sgd_iter: 32
|
|
|
|
|
clip_param: 0.2
|
|
|
|
|
lr_schedule: [
|
|
|
|
|
[0, 0.0005],
|
|
|
|
|
[150000000, 0.000001],
|
|
|
|
|
]
|
|
|
|
|
vf_loss_coeff: 0.5
|
|
|
|
|
entropy_coeff: 0.01
|
|
|
|
|
grad_clip: 0.5
|
|
|
|
|
batch_mode: truncate_episodes
|
|
|
|
|
use_kl_loss: True
|
|
|
|
|
kl_coeff: 1.0
|
|
|
|
|
kl_target: 0.04
|
|
|
|
|
observation_filter: MeanStdFilter
|
|
|
|
|
|
