Update config

Skylark0924 · Aug 24, 2023 · e5c3428 · e5c3428
1 parent 147a41e
commit e5c3428
Show file tree

Hide file tree

Showing 16 changed files with 48 additions and 0 deletions.
diff --git a/rofunc/config/learning/rl/train/AntPPORofuncRL.yaml b/rofunc/config/learning/rl/train/AntPPORofuncRL.yaml
@@ -16,7 +16,10 @@ Trainer:
   start_learning_steps: 0             # The number of steps to take before starting network updating.
   seed: 42                            # The random seed.
   rollouts: 16                        # The number of rollouts before updating.
+  eval_flag: False                     # If true, run evaluation.
+  eval_freq: 2500                     # The frequency of evaluation. (timesteps)
   eval_steps: 1000                    # The number of steps to run for evaluation.
+  use_eval_thread: True               # If true, use a separate thread for evaluation.
   inference_steps: 1000               # The number of steps to run for inference.
   max_episode_steps: 200              # The maximum number of steps per episode.
 

diff --git a/rofunc/config/learning/rl/train/BaseTaskA2CRofuncRL.yaml b/rofunc/config/learning/rl/train/BaseTaskA2CRofuncRL.yaml
@@ -14,7 +14,10 @@ Trainer:
   start_learning_steps: 0             # The number of steps to take before starting network updating.
   seed: 42                            # The random seed.
   rollouts: 16                        # The number of rollouts before updating.
+  eval_flag: False                     # If true, run evaluation.
+  eval_freq: 2500                     # The frequency of evaluation. (timesteps)
   eval_steps: 1000                    # The number of steps to run for evaluation.
+  use_eval_thread: True               # If true, use a separate thread for evaluation.
   inference_steps: 1000               # The number of steps to run for inference.
   max_episode_steps: 200              # The maximum number of steps per episode.
 

diff --git a/rofunc/config/learning/rl/train/BaseTaskAMPRofuncRL.yaml b/rofunc/config/learning/rl/train/BaseTaskAMPRofuncRL.yaml
@@ -14,7 +14,10 @@ Trainer:
   start_learning_steps: 0             # The number of steps to take before starting network updating.
   seed: 42                            # The random seed.
   rollouts: 16                        # The number of rollouts before updating.
+  eval_flag: False                     # If true, run evaluation.
+  eval_freq: 2500                     # The frequency of evaluation. (timesteps)
   eval_steps: 1000                    # The number of steps to run for evaluation.
+  use_eval_thread: True               # If true, use a separate thread for evaluation.
   inference_steps: 1000               # The number of steps to run for inference.
   max_episode_steps: 200              # The maximum number of steps per episode.
 

diff --git a/rofunc/config/learning/rl/train/BaseTaskPPORofuncRL.yaml b/rofunc/config/learning/rl/train/BaseTaskPPORofuncRL.yaml
@@ -14,7 +14,10 @@ Trainer:
   start_learning_steps: 0             # The number of steps to take before starting network updating.
   seed: 42                            # The random seed.
   rollouts: 16                        # The number of rollouts before updating.
+  eval_flag: True                     # If true, run evaluation.
+  eval_freq: 2500                     # The frequency of evaluation. (timesteps)
   eval_steps: 1000                    # The number of steps to run for evaluation.
+  use_eval_thread: True               # If true, use a separate thread for evaluation.
   inference_steps: 1000               # The number of steps to run for inference.
   max_episode_steps: 200              # The maximum number of steps per episode.
 

diff --git a/rofunc/config/learning/rl/train/BaseTaskSACRofuncRL.yaml b/rofunc/config/learning/rl/train/BaseTaskSACRofuncRL.yaml
@@ -14,7 +14,10 @@ Trainer:
   start_learning_steps: 1000             # The number of steps to take before starting network updating.
   seed: 42                            # The random seed.
   rollouts: 16                        # The number of rollouts before updating.
+  eval_flag: False                     # If true, run evaluation.
+  eval_freq: 2500                     # The frequency of evaluation. (timesteps)
   eval_steps: 1000                    # The number of steps to run for evaluation.
+  use_eval_thread: True               # If true, use a separate thread for evaluation.
   inference_steps: 1000               # The number of steps to run for inference.
   max_episode_steps: 200              # The maximum number of steps per episode.
 

diff --git a/rofunc/config/learning/rl/train/BaseTaskTD3RofuncRL.yaml b/rofunc/config/learning/rl/train/BaseTaskTD3RofuncRL.yaml
@@ -14,7 +14,10 @@ Trainer:
   start_learning_steps: 0             # The number of steps to take before starting network updating.
   seed: 42                            # The random seed.
   rollouts: 16                        # The number of rollouts before updating.
+  eval_flag: False                     # If true, run evaluation.
+  eval_freq: 2500                     # The frequency of evaluation. (timesteps)
   eval_steps: 1000                    # The number of steps to run for evaluation.
+  use_eval_thread: True               # If true, use a separate thread for evaluation.
   inference_steps: 1000               # The number of steps to run for inference.
   max_episode_steps: 200              # The maximum number of steps per episode.
 

diff --git a/rofunc/config/learning/rl/train/CURICabinetImagePPORofuncRL.yaml b/rofunc/config/learning/rl/train/CURICabinetImagePPORofuncRL.yaml
@@ -16,7 +16,10 @@ Trainer:
   start_learning_steps: 0             # The number of steps to take before starting network updating.
   seed: 42                            # The random seed.
   rollouts: 16                        # The number of rollouts before updating.
+  eval_flag: False                     # If true, run evaluation.
+  eval_freq: 2500                     # The frequency of evaluation. (timesteps)
   eval_steps: 1000                    # The number of steps to run for evaluation.
+  use_eval_thread: True               # If true, use a separate thread for evaluation.
   inference_steps: 1000               # The number of steps to run for inference.
   max_episode_steps: 200              # The maximum number of steps per episode.
 

diff --git a/rofunc/config/learning/rl/train/CURICabinetPPORofuncRL.yaml b/rofunc/config/learning/rl/train/CURICabinetPPORofuncRL.yaml
@@ -16,7 +16,10 @@ Trainer:
   start_learning_steps: 0             # The number of steps to take before starting network updating.
   seed: 42                            # The random seed.
   rollouts: 16                        # The number of rollouts before updating.
+  eval_flag: False                     # If true, run evaluation.
+  eval_freq: 2500                     # The frequency of evaluation. (timesteps)
   eval_steps: 1000                    # The number of steps to run for evaluation.
+  use_eval_thread: True               # If true, use a separate thread for evaluation.
   inference_steps: 1000               # The number of steps to run for inference.
   max_episode_steps: 200              # The maximum number of steps per episode.
 

diff --git a/rofunc/config/learning/rl/train/CURICabinetSACRofuncRL.yaml b/rofunc/config/learning/rl/train/CURICabinetSACRofuncRL.yaml
@@ -15,7 +15,10 @@ Trainer:
   start_learning_steps: 0             # The number of steps to take before starting network updating.
   seed: 42                            # The random seed.
   rollouts: 16                        # The number of rollouts before updating.
+  eval_flag: False                     # If true, run evaluation.
+  eval_freq: 2500                     # The frequency of evaluation. (timesteps)
   eval_steps: 1000                    # The number of steps to run for evaluation.
+  use_eval_thread: True               # If true, use a separate thread for evaluation.
   inference_steps: 1000               # The number of steps to run for inference.
   max_episode_steps: 200              # The maximum number of steps per episode.
 

diff --git a/rofunc/config/learning/rl/train/FrankaCabinetPPORofuncRL.yaml b/rofunc/config/learning/rl/train/FrankaCabinetPPORofuncRL.yaml
@@ -16,7 +16,10 @@ Trainer:
   start_learning_steps: 0             # The number of steps to take before starting network updating.
   seed: 42                            # The random seed.
   rollouts: 16                        # The number of rollouts before updating.
+  eval_flag: False                     # If true, run evaluation.
+  eval_freq: 2500                     # The frequency of evaluation. (timesteps)
   eval_steps: 1000                    # The number of steps to run for evaluation.
+  use_eval_thread: True               # If true, use a separate thread for evaluation.
   inference_steps: 1000               # The number of steps to run for inference.
   max_episode_steps: 200              # The maximum number of steps per episode.
 

diff --git a/rofunc/config/learning/rl/train/HumanoidASEGetupSwordShieldASERofuncRL.yaml b/rofunc/config/learning/rl/train/HumanoidASEGetupSwordShieldASERofuncRL.yaml
@@ -14,7 +14,10 @@ Trainer:
   start_learning_steps: 0             # The number of steps to take before starting network updating.
   seed: 42                            # The random seed.
   rollouts: 32                        # The number of rollouts before updating.
+  eval_flag: False                     # If true, run evaluation.
+  eval_freq: 2500                     # The frequency of evaluation. (timesteps)
   eval_steps: 1000                    # The number of steps to run for evaluation.
+  use_eval_thread: True               # If true, use a separate thread for evaluation.
   inference_steps: 1000               # The number of steps to run for inference.
   max_episode_steps: 200              # The maximum number of steps per episode.
 

diff --git a/rofunc/config/learning/rl/train/HumanoidASEHeadingSwordShieldASERofuncRL.yaml b/rofunc/config/learning/rl/train/HumanoidASEHeadingSwordShieldASERofuncRL.yaml
@@ -14,7 +14,10 @@ Trainer:
   start_learning_steps: 0             # The number of steps to take before starting network updating.
   seed: 42                            # The random seed.
   rollouts: 32                        # The number of rollouts before updating.
+  eval_flag: False                     # If true, run evaluation.
+  eval_freq: 2500                     # The frequency of evaluation. (timesteps)
   eval_steps: 1000                    # The number of steps to run for evaluation.
+  use_eval_thread: True               # If true, use a separate thread for evaluation.
   inference_steps: 1000               # The number of steps to run for inference.
   max_episode_steps: 200              # The maximum number of steps per episode.
 

diff --git a/rofunc/config/learning/rl/train/HumanoidASELocationSwordShieldASERofuncRL.yaml b/rofunc/config/learning/rl/train/HumanoidASELocationSwordShieldASERofuncRL.yaml
@@ -14,7 +14,10 @@ Trainer:
   start_learning_steps: 0             # The number of steps to take before starting network updating.
   seed: 42                            # The random seed.
   rollouts: 32                        # The number of rollouts before updating.
+  eval_flag: False                     # If true, run evaluation.
+  eval_freq: 2500                     # The frequency of evaluation. (timesteps)
   eval_steps: 1000                    # The number of steps to run for evaluation.
+  use_eval_thread: True               # If true, use a separate thread for evaluation.
   inference_steps: 1000               # The number of steps to run for inference.
   max_episode_steps: 200              # The maximum number of steps per episode.
 

diff --git a/rofunc/config/learning/rl/train/HumanoidASEPerturbSwordShieldASERofuncRL.yaml b/rofunc/config/learning/rl/train/HumanoidASEPerturbSwordShieldASERofuncRL.yaml
@@ -14,7 +14,10 @@ Trainer:
   start_learning_steps: 0             # The number of steps to take before starting network updating.
   seed: 42                            # The random seed.
   rollouts: 32                        # The number of rollouts before updating.
+  eval_flag: False                     # If true, run evaluation.
+  eval_freq: 2500                     # The frequency of evaluation. (timesteps)
   eval_steps: 1000                    # The number of steps to run for evaluation.
+  use_eval_thread: True               # If true, use a separate thread for evaluation.
   inference_steps: 1000               # The number of steps to run for inference.
   max_episode_steps: 200              # The maximum number of steps per episode.
 

diff --git a/rofunc/config/learning/rl/train/HumanoidASEReachSwordShieldASERofuncRL.yaml b/rofunc/config/learning/rl/train/HumanoidASEReachSwordShieldASERofuncRL.yaml
@@ -14,7 +14,10 @@ Trainer:
   start_learning_steps: 0             # The number of steps to take before starting network updating.
   seed: 42                            # The random seed.
   rollouts: 32                        # The number of rollouts before updating.
+  eval_flag: False                     # If true, run evaluation.
+  eval_freq: 2500                     # The frequency of evaluation. (timesteps)
   eval_steps: 1000                    # The number of steps to run for evaluation.
+  use_eval_thread: True               # If true, use a separate thread for evaluation.
   inference_steps: 1000               # The number of steps to run for inference.
   max_episode_steps: 200              # The maximum number of steps per episode.
 

diff --git a/rofunc/config/learning/rl/train/HumanoidASEStrikeSwordShieldASERofuncRL.yaml b/rofunc/config/learning/rl/train/HumanoidASEStrikeSwordShieldASERofuncRL.yaml
@@ -14,7 +14,10 @@ Trainer:
   start_learning_steps: 0             # The number of steps to take before starting network updating.
   seed: 42                            # The random seed.
   rollouts: 32                        # The number of rollouts before updating.
+  eval_flag: False                     # If true, run evaluation.
+  eval_freq: 2500                     # The frequency of evaluation. (timesteps)
   eval_steps: 1000                    # The number of steps to run for evaluation.
+  use_eval_thread: True               # If true, use a separate thread for evaluation.
   inference_steps: 1000               # The number of steps to run for inference.
   max_episode_steps: 200              # The maximum number of steps per episode.