learning_rate_schedule: linear
encoder_layers: 2
policy_layers: 2
value_layers: 3
value_layers: 2
feature_size: 128
in_epoch_alter: true
use_op_buffer: true
use_bisim: true
separate_value_net: true
# separate_value_net: true
network_settings:
normalize: true
hidden_units: 512
use_transfer: true
load_policy: false
load_value: false