action_layers: 2
policy_layers: 2
forward_layers: 0
value_layers: 3
value_layers: 2
feature_size: 128
action_feature_size: 128
reuse_encoder: true
predict_return: true
use_bisim: false
separate_value_train: true
separate_value_net: true
in_batch_alter: true
network_settings:
normalize: true
reuse_encoder: false
train_model: false
load_model: true
train_action: false