files organize

examples/ant_ppo._ms.yaml

@@ -0,0 +1,43 @@
+alg_para:
+ alg_name: PPO
+
+env_para:
+ env_name: GymEnv
+ env_info:
+ name: MiniGrid-Ant-v0
+ vision: False
+
+agent_para:
+ agent_name: AtariPpo
+ agent_num : 1
+ agent_config:
+ max_steps: 200
+ complete_step: 100000
+
+model_para:
+ actor:
+ model_name: PpoCnnMS
+ state_dim: [84, 84, 3]
+ action_dim: 4
+ input_dtype: uint8
+ model_config:
+ BATCH_SIZE: 10
+ CRITIC_LOSS_COEF: 1.0
+ ENTROPY_LOSS: 0.003
+ LOSS_CLIPPING: 0.1
+ LR: 0.00025
+ MAX_GRAD_NORM: 5.0
+ NUM_SGD_ITER: 4
+ SUMMARY: False
+ VF_SHARE_LAYERS: True
+ activation: relu
+ hidden_sizes: [512]
+
+
+env_num: 1
+
+benchmark:
+ log_interval_to_train: 5
+
+
+

examples/beamrider_dqn_ms.yaml

@@ -0,0 +1,32 @@
+alg_para:
+ alg_name: DQN
+ alg_config: {
+ 'train_per_checkpoint': 50,
+ 'prepare_times_per_train': 4,
+ 'learning_starts': 10000,
+ 'BUFFER_SIZE': 400000,
+ }
+
+env_para:
+ env_name: AtariEnv
+ env_info: { 'name': BeamRiderNoFrameskip-v4, 'vision': False}
+
+agent_para:
+ agent_name: AtariDqn
+ agent_num : 1
+ agent_config: {
+ 'max_steps': 2000,
+ 'complete_step': 10000000,
+ 'episode_count': 200000
+ }
+
+model_para:
+ actor:
+ model_name: DqnCnnMS
+ state_dim: [84,84,4]
+ action_dim: 9
+ model_config: {
+ 'LR': 0.00015,
+ }
+
+env_num: 2

examples/beamrider_ppo_ms.yaml

@@ -0,0 +1,39 @@
+alg_para:
+ alg_name: PPO
+
+env_para:
+ env_name: AtariEnv
+ env_info:
+ name: BeamRiderNoFrameskip-v4
+ vision: False
+
+agent_para:
+ agent_name: AtariPpo
+ agent_num : 1
+ agent_config:
+ max_steps: 128
+ complete_step: 10000000
+
+model_para:
+ actor:
+ model_name: PpoCnnMS
+ state_dim: [84, 84, 4]
+ action_dim: 9
+ input_dtype: uint8
+ model_config:
+ BATCH_SIZE: 320
+ CRITIC_LOSS_COEF: 1.0
+ ENTROPY_LOSS: 0.003
+ LOSS_CLIPPING: 0.1
+ LR: 0.00025
+ MAX_GRAD_NORM: 5.0
+ NUM_SGD_ITER: 4
+ SUMMARY: False
+ VF_SHARE_LAYERS: True
+ activation: relu
+ hidden_sizes: [512]
+
+env_num: 10
+
+benchmark:
+ log_interval_to_train: 10

examples/breakout_dqn_ms.yaml

@@ -0,0 +1,32 @@
+alg_para:
+ alg_name: DQN
+ alg_config: {
+ 'train_per_checkpoint': 50,
+ 'prepare_times_per_train': 4,
+ 'learning_starts': 10000,
+ 'BUFFER_SIZE': 400000,
+ }
+
+env_para:
+ env_name: AtariEnv
+ env_info: { 'name': BreakoutNoFrameskip-v4, 'vision': False}
+
+agent_para:
+ agent_name: AtariDqn
+ agent_num : 1
+ agent_config: {
+ 'max_steps': 2000,
+ 'complete_step': 10000000,
+ 'episode_count': 200000
+ }
+
+model_para:
+ actor:
+ model_name: DqnCnnMS
+ state_dim: [84,84,4]
+ action_dim: 4
+ model_config: {
+ 'LR': 0.00015,
+ }
+
+env_num: 2

examples/breakout_ppo_ms.yaml

@@ -0,0 +1,40 @@
+alg_para:
+ alg_name: PPO
+
+env_para:
+ env_name: AtariEnv
+ env_info:
+ name: BreakoutNoFrameskip-v4
+ vision: False
+
+agent_para:
+ agent_name: AtariPpo
+ agent_num : 1
+ agent_config:
+ max_steps: 128
+ complete_step: 10000000
+
+model_para:
+ actor:
+ model_name: PpoCnnMS
+ state_dim: [84, 84, 4]
+ action_dim: 4
+ input_dtype: uint8
+ model_config:
+ BATCH_SIZE: 320
+ CRITIC_LOSS_COEF: 1.0
+ ENTROPY_LOSS: 0.003
+ LOSS_CLIPPING: 0.1
+ LR: 0.00025
+ MAX_GRAD_NORM: 5.0
+ NUM_SGD_ITER: 4
+ SUMMARY: False
+ VF_SHARE_LAYERS: True
+ activation: relu
+ hidden_sizes: [256]
+
+env_num: 10
+speedup: False
+
+benchmark:
+ log_interval_to_train: 10

examples/cartpole_dqn_ms.yaml

@@ -0,0 +1,29 @@
+alg_para:
+ alg_name: DQN
+ alg_config: {
+ 'train_per_checkpoint': 10,
+ 'prepare_times_per_train': 4,
+ 'learning_starts': 100,
+ 'save_model': True,
+ 'save_interval': 100
+ }
+
+env_para:
+ env_name: GymEnv
+ env_info: { 'name': CartPole-v0, 'vision': False}
+
+agent_para:
+ agent_name: CartpoleDqn
+ agent_num : 1
+ agent_config: {
+ 'max_steps': 2000 ,
+ 'complete_step': 5000000
+ }
+
+model_para:
+ actor:
+ model_name: DqnMlpMS
+ state_dim: [4]
+ action_dim: 2
+
+env_num: 1

examples/cartpole_ppo_ms.yaml

@@ -0,0 +1,52 @@
+alg_para:
+ alg_name: PPO
+ alg_config:
+ process_num: 1
+ save_model: True # default False
+ save_interval: 100
+
+env_para:
+ env_name: GymEnv
+ env_info:
+ name: CartPole-v0
+ vision: False
+
+agent_para:
+ agent_name: PPO
+ agent_num : 1
+ agent_config:
+ max_steps: 200
+ complete_step: 2000000
+ complete_episode: 9000
+
+model_para:
+ actor:
+ model_name: PpoMlpMS
+ state_dim: [4]
+ action_dim: 2
+ input_dtype: float32
+ model_config:
+ BATCH_SIZE: 200
+ CRITIC_LOSS_COEF: 1.0
+ ENTROPY_LOSS: 0.01
+ LR: 0.0003
+ LOSS_CLIPPING: 0.2
+ MAX_GRAD_NORM: 5.0
+ NUM_SGD_ITER: 8
+ SUMMARY: False
+ VF_SHARE_LAYERS: False
+ activation: tanh
+ hidden_sizes: [64, 64]
+
+env_num: 10
+speedup: False
+
+benchmark:
+ log_interval_to_train: 20
+ eval:
+ # model_path: /YOUR/PATH/TO/EVAL/models
+ gap: 100
+ model_divided_freq: 1 # how many times to divided of each model
+ episodes_per_eval: 1
+ evaluator_num: 1
+ max_step_per_episode: 4000

examples/dog_ppo_ms.yaml

@@ -0,0 +1,43 @@
+alg_para:
+ alg_name: PPO
+
+env_para:
+ env_name: GymEnv
+ env_info:
+ name: MiniGrid-Dog-v0
+ vision: False
+
+agent_para:
+ agent_name: AtariPpo
+ agent_num : 1
+ agent_config:
+ max_steps: 200
+ complete_step: 100000
+
+model_para:
+ actor:
+ model_name: PpoCnnMS
+ state_dim: [84, 84, 3]
+ action_dim: 4
+ input_dtype: uint8
+ model_config:
+ BATCH_SIZE: 10
+ CRITIC_LOSS_COEF: 1.0
+ ENTROPY_LOSS: 0.003
+ LOSS_CLIPPING: 0.1
+ LR: 0.00025
+ MAX_GRAD_NORM: 5.0
+ NUM_SGD_ITER: 4
+ SUMMARY: False
+ VF_SHARE_LAYERS: True
+ activation: relu
+ hidden_sizes: [512]
+
+
+env_num: 1
+
+benchmark:
+ log_interval_to_train: 5
+
+
+

examples/pendulum_ppo_ms.yaml

@@ -0,0 +1,42 @@
+alg_para:
+ alg_name: PPO
+ alg_config:
+ process_num: 1
+ only_save_best_model: True
+
+env_para:
+ env_name: GymEnv
+ env_info:
+ name: Pendulum-v0
+ vision: False
+
+agent_para:
+ agent_name: PPO
+ agent_num : 1
+ agent_config:
+ max_steps: 200
+ complete_step: 2000000
+
+model_para:
+ actor:
+ model_name: PpoMlpMS
+ state_dim: [3]
+ action_dim: 1
+ input_dtype: float32
+ model_config:
+ BATCH_SIZE: 200
+ CRITIC_LOSS_COEF: 1.0
+ ENTROPY_LOSS: 0.01
+ LR: 0.0003
+ LOSS_CLIPPING: 0.2
+ MAX_GRAD_NORM: 5.0
+ NUM_SGD_ITER: 8
+ SUMMARY: False
+ VF_SHARE_LAYERS: False
+ activation: tanh
+ hidden_sizes: [64, 64]
+
+env_num: 10
+
+benchmark:
+ log_interval_to_train: 10

examples/pong_dqn_ms.yaml

@@ -0,0 +1,30 @@
+alg_para:
+ alg_name: DQN
+ alg_config: {
+ 'train_per_checkpoint': 50,
+ 'prepare_times_per_train': 4,
+ 'learning_starts': 10000,
+ 'BUFFER_SIZE': 10000,
+# 'save_model': True,
+ 'save_interval': 100
+ }
+
+env_para:
+ env_name: AtariEnv
+ env_info: { 'name': PongNoFrameskip-v4, 'vision': False}
+
+agent_para:
+ agent_name: AtariDqn
+ agent_num : 1
+ agent_config: {
+ 'max_steps': 2000,
+ 'complete_step': 10000000
+ }
+
+model_para:
+ actor:
+ model_name: DqnCnnPongMS
+ state_dim: [84,84,4]
+ action_dim: 6
+
+env_num: 2