forward_layers: 1
value_layers: 1
feature_size: 16
separate_value_net: true
# separate_value_net: true
separate_policy_train: true
reuse_encoder: true
in_epoch_alter: false
in_batch_alter: true
predict_return: true
use_bisim: false
use_bisim: true
network_settings:
normalize: true
hidden_units: 64
reuse_encoder: false
in_batch_alter: false
transfer_path: "results/sac_model_ball/3DBall"
transfer_path: "results/sac_model_ball_bisim/3DBall"
self.sequence_length_ph = self.policy.sequence_length_ph
hidden_critic = self._create_encoder(
self.visual_in,
self.processed_vector_in,
self.policy.visual_in,
self.policy.processed_vector_in,
vis_encode_type,
encoder_layers=encoder_layers,
scope="encoding",