76 lines
1.3 KiB
YAML
76 lines
1.3 KiB
YAML
|
dmc:
|
||
|
|
|
||
|
|
logdir: ./
|
||
|
|
video_dir: ./
|
||
|
|
debug: False
|
||
|
|
seed: 0
|
||
|
|
steps: 1000000.0
|
||
|
|
snapshot_every: 20000.0
|
||
|
|
eval_every: 5000.0
|
||
|
|
log_every: 5000.0
|
||
|
|
image_size: 64
|
||
|
|
log_scalars: true
|
||
|
|
log_images: true
|
||
|
|
gpu_growth: true
|
||
|
|
precision: 16
|
||
|
|
task: dmc_cheetah_run_driving
|
||
|
|
envs: 1
|
||
|
|
parallel: none
|
||
|
|
action_repeat: 2
|
||
|
|
time_limit: 1000
|
||
|
|
prefill: 5000
|
||
|
|
eval_noise: 0.0
|
||
|
|
clip_rewards: none
|
||
|
|
deter_size: 200
|
||
|
|
stoch_size: 30
|
||
|
|
disen_deter_size: 200
|
||
|
|
disen_stoch_size: 30
|
||
|
|
num_units: 400
|
||
|
|
dense_act: elu
|
||
|
|
cnn_act: relu
|
||
|
|
cnn_depth: 26
|
||
|
|
disen_cnn_depth: 26
|
||
|
|
pcont: false
|
||
|
|
free_nats: 3.0
|
||
|
|
num_reward_opt_iters: 20
|
||
|
|
disen_neg_rew_scale: 20000.0
|
||
|
|
disen_rec_scale: 1.5
|
||
|
|
disen_kl_scale: 1.0
|
||
|
|
kl_scale: 1.0
|
||
|
|
reward_scale: 1.0
|
||
|
|
pcont_scale: 10.0
|
||
|
|
weight_decay: 0.0
|
||
|
|
weight_decay_pattern: .*
|
||
|
|
batch_size: 50
|
||
|
|
batch_length: 50
|
||
|
|
train_every: 1000
|
||
|
|
train_steps: 100
|
||
|
|
pretrain: 100
|
||
|
|
disen_reward_lr: 0.0006
|
||
|
|
model_lr: 0.0006
|
||
|
|
value_lr: 8.0e-05
|
||
|
|
actor_lr: 8.0e-05
|
||
|
|
grad_clip: 100.0
|
||
|
|
dataset_balance: false
|
||
|
|
discount: 0.99
|
||
|
|
disclam: 0.95
|
||
|
|
horizon: 15
|
||
|
|
action_dist: tanh_normal
|
||
|
|
action_init_std: 5.0
|
||
|
|
expl: additive_gaussian
|
||
|
|
expl_amount: 0.3
|
||
|
|
expl_decay: 0.0
|
||
|
|
expl_min: 0.0
|
||
|
|
|
||
|
|
debug:
|
||
|
|
|
||
|
|
debug: True
|
||
|
|
pretrain: 1
|
||
|
|
prefill: 1
|
||
|
|
train_steps: 1
|
||
|
|
batch_size: 10
|
||
|
|
batch_length: 20
|
||
|
|
|
||
|
|
|
||
|
|
|