cloud config: non-linear policy

config/sac_transfer/3DBallCloud.yaml

+behaviors:
+  3DBall:
+    trainer_type: sac_transfer
+    hyperparameters:
+      learning_rate: 0.0003
+      learning_rate_schedule: linear
+      model_schedule: linear
+      batch_size: 64
+      buffer_size: 200000
+      buffer_init_steps: 0
+      tau: 0.005
+      steps_per_update: 10.0
+      save_replay_buffer: false
+      init_entcoef: 0.5
+      reward_signal_steps_per_update: 10.0
+      encoder_layers: 2
+      policy_layers: 1
+      forward_layers: 0
+      value_layers: 1
+      feature_size: 32
+      # separate_value_net: true
+      separate_policy_train: true
+      # separate_value_train: true
+      separate_model_train: true
+      reuse_encoder: true
+      in_epoch_alter: false
+      in_batch_alter: true
+      use_op_buffer: false
+      use_var_predict: true
+      with_prior: false
+      predict_return: true
+      use_bisim: false
+    network_settings:
+      normalize: true
+      hidden_units: 128
+      num_layers: 2
+      vis_encode_type: simple
+    reward_signals:
+      extrinsic:
+        gamma: 0.99
+        strength: 1.0
+    keep_checkpoints: 5
+    max_steps: 500000
+    time_horizon: 1000
+    summary_freq: 12000
+    threaded: true

config/sac_transfer/3DBallHardCloud.yaml

+behaviors:
+  3DBallHard:
+    trainer_type: sac_transfer
+    hyperparameters:
+      learning_rate: 0.0003
+      learning_rate_schedule: linear
+      batch_size: 256
+      buffer_size: 500000
+      buffer_init_steps: 0
+      tau: 0.005
+      steps_per_update: 10.0
+      save_replay_buffer: false
+      init_entcoef: 1.0
+      reward_signal_steps_per_update: 10.0
+      encoder_layers: 2
+      policy_layers: 1
+      forward_layers: 0
+      value_layers: 1
+      feature_size: 32
+      # separate_value_net: true
+      separate_policy_train: true
+      # separate_value_train: true
+      separate_model_train: true
+      reuse_encoder: true
+      in_epoch_alter: false
+      in_batch_alter: true
+      use_op_buffer: false
+      use_var_predict: true
+      with_prior: false
+      predict_return: true
+      use_bisim: false
+    network_settings:
+      normalize: true
+      hidden_units: 128
+      num_layers: 2
+      vis_encode_type: simple
+    reward_signals:
+      extrinsic:
+        gamma: 0.99
+        strength: 1.0
+    keep_checkpoints: 5
+    max_steps: 500000
+    time_horizon: 1000
+    summary_freq: 12000
+    threaded: true

config/sac_transfer/3DBallHardTransferCloud.yaml

+behaviors:
+  3DBallHard:
+    trainer_type: sac_transfer
+    hyperparameters:
+      learning_rate: 0.0003
+      learning_rate_schedule: linear
+      batch_size: 256
+      buffer_size: 500000
+      buffer_init_steps: 0
+      tau: 0.005
+      steps_per_update: 10.0
+      save_replay_buffer: false
+      init_entcoef: 1.0
+      reward_signal_steps_per_update: 10.0
+      encoder_layers: 2
+      policy_layers: 1
+      forward_layers: 0
+      value_layers: 1
+      feature_size: 32
+      # separate_value_net: true
+      separate_policy_train: true
+      # separate_value_train: true
+      reuse_encoder: true
+      in_epoch_alter: false
+      in_batch_alter: false
+      use_op_buffer: false
+      use_var_predict: true
+      with_prior: false
+      predict_return: true
+      use_bisim: false
+      use_transfer: true
+      load_model: true
+      train_model: false
+      transfer_path: "results/sac-ball-f32-p1f0/3DBall"
+    network_settings:
+      normalize: true
+      hidden_units: 128
+      num_layers: 2
+      vis_encode_type: simple
+    reward_signals:
+      extrinsic:
+        gamma: 0.99
+        strength: 1.0
+    keep_checkpoints: 5
+    max_steps: 500000
+    time_horizon: 1000
+    summary_freq: 12000
+    threaded: true