mirror of
https://github.com/facebookresearch/ReAgent.git
synced 2026-05-17 12:40:39 +00:00
Jason Gauci
39385e8d83
Summary: Pull Request resolved: https://github.com/facebookresearch/ReAgent/pull/470 Reviewed By: czxttkl Differential Revision: D28093192 fbshipit-source-id: 6b260c3e8d49c8b302e40066e2be49a0bfe96688
59 lines
1.2 KiB
YAML
59 lines
1.2 KiB
YAML
env:
|
|
Gym:
|
|
env_name: Pendulum-v0
|
|
model:
|
|
SAC:
|
|
trainer_param:
|
|
rl:
|
|
gamma: 0.99
|
|
target_update_rate: 0.005
|
|
softmax_policy: true
|
|
entropy_temperature: 0.3
|
|
crr_config:
|
|
exponent_beta: 1.0
|
|
exponent_clamp: 20.0
|
|
q_network_optimizer:
|
|
Adam:
|
|
lr: 0.001
|
|
value_network_optimizer:
|
|
Adam:
|
|
lr: 0.001
|
|
actor_network_optimizer:
|
|
Adam:
|
|
lr: 0.001
|
|
actor_net_builder:
|
|
GaussianFullyConnected:
|
|
sizes:
|
|
- 64
|
|
- 64
|
|
activations:
|
|
- leaky_relu
|
|
- leaky_relu
|
|
critic_net_builder:
|
|
FullyConnected:
|
|
sizes:
|
|
- 64
|
|
- 64
|
|
activations:
|
|
- leaky_relu
|
|
- leaky_relu
|
|
value_net_builder:
|
|
FullyConnected:
|
|
sizes:
|
|
- 64
|
|
- 64
|
|
activations:
|
|
- leaky_relu
|
|
- leaky_relu
|
|
eval_parameters:
|
|
calc_cpe_in_training: false
|
|
replay_memory_size: 100000
|
|
train_every_ts: 1
|
|
train_after_ts: 10000
|
|
num_train_episodes: 40
|
|
num_eval_episodes: 20
|
|
# Though maximal score is 0, we set lower bar to let tests finish in time
|
|
passing_score_bar: -500
|
|
use_gpu: false
|
|
minibatch_size: 256
|