slm_lab/spec/experimental/ppo/ppo_pendulum.json
{
"ppo_shared_pendulum": {
"agent": [{
"name": "PPO",
"algorithm": {
"name": "PPO",
"action_pdtype": "default",
"action_policy": "default",
"explore_var_spec": null,
"gamma": 0.99,
"lam": 0.95,
"clip_eps_spec": {
"name": "linear_decay",
"start_val": 0.10,
"end_val": 0.01,
"start_step": 1000,
"end_step": 5000,
},
"entropy_coef_spec": {
"name": "linear_decay",
"start_val": 0.01,
"end_val": 0.001,
"start_step": 1000,
"end_step": 5000,
},
"val_loss_coef": 0.1,
"time_horizon": 8,
"training_epoch": 8
},
"memory": {
"name": "OnPolicyBatchReplay"
},
"net": {
"type": "MLPNet",
"shared": true,
"hid_layers": [32],
"hid_layers_activation": "relu",
"clip_grad_val": 10.0,
"use_same_optim": false,
"actor_optim_spec": {
"name": "Adam",
"lr": 0.02
},
"critic_optim_spec": {
"name": "Adam",
"lr": 0.02
},
"lr_scheduler_spec": {
"name": "StepLR",
"step_size": 2000,
"gamma": 0.9,
},
"gpu": false
}
}],
"env": [{
"name": "Pendulum-v0",
"max_t": null,
"max_frame": 500,
}],
"body": {
"product": "outer",
"num": 1
},
"meta": {
"distributed": false,
"eval_frequency": 1000,
"max_session": 4,
"max_trial": 100,
"search": "RandomSearch"
},
"search": {
"agent": [{
"net": {
"hid_layers__choice": [
[32],
[64],
[32, 16],
[64, 32]
],
"hid_layers_activation__choice": ["sigmoid", "relu", "tanh"],
"actor_optim_spec": {
"lr__uniform": [0.0001, 0.2]
},
"critic_optim_spec": {
"lr__uniform": [0.0001, 0.2]
}
}
}]
}
},
"ppo_separate_pendulum": {
"agent": [{
"name": "PPO",
"algorithm": {
"name": "PPO",
"action_pdtype": "default",
"action_policy": "default",
"explore_var_spec": null,
"gamma": 0.99,
"lam": 0.95,
"clip_eps_spec": {
"name": "linear_decay",
"start_val": 0.10,
"end_val": 0.01,
"start_step": 1000,
"end_step": 5000,
},
"entropy_coef_spec": {
"name": "linear_decay",
"start_val": 0.01,
"end_val": 0.001,
"start_step": 1000,
"end_step": 5000,
},
"val_loss_coef": 0.85,
"time_horizon": 8,
"training_epoch": 8
},
"memory": {
"name": "OnPolicyBatchReplay"
},
"net": {
"type": "MLPNet",
"shared": false,
"hid_layers": [64],
"hid_layers_activation": "tanh",
"clip_grad_val": 10.0,
"use_same_optim": false,
"actor_optim_spec": {
"name": "Adam",
"lr": 0.01
},
"critic_optim_spec": {
"name": "Adam",
"lr": 0.05
},
"lr_scheduler_spec": {
"name": "StepLR",
"step_size": 2000,
"gamma": 0.9,
},
"gpu": false
}
}],
"env": [{
"name": "Pendulum-v0",
"max_t": null,
"max_frame": 500,
}],
"body": {
"product": "outer",
"num": 1
},
"meta": {
"distributed": false,
"eval_frequency": 1000,
"max_session": 4,
"max_trial": 100,
"search": "RandomSearch"
},
"search": {
"agent": [{
"net": {
"hid_layers__choice": [
[32],
[64],
[32, 16],
[64, 32]
],
"hid_layers_activation__choice": ["sigmoid", "relu", "tanh"],
"actor_optim_spec": {
"lr__uniform": [0.0001, 0.2]
},
"critic_optim_spec": {
"lr__uniform": [0.0001, 0.2]
}
}
}]
}
},
"ppo_rnn_shared_pendulum": {
"agent": [{
"name": "PPO",
"algorithm": {
"name": "PPO",
"action_pdtype": "default",
"action_policy": "default",
"explore_var_spec": null,
"gamma": 0.99,
"lam": 0.95,
"clip_eps_spec": {
"name": "linear_decay",
"start_val": 0.10,
"end_val": 0.01,
"start_step": 1000,
"end_step": 5000,
},
"entropy_coef_spec": {
"name": "linear_decay",
"start_val": 0.01,
"end_val": 0.001,
"start_step": 1000,
"end_step": 5000,
},
"val_loss_coef": 0.1,
"time_horizon": 8,
"training_epoch": 8
},
"memory": {
"name": "OnPolicyBatchReplay"
},
"net": {
"type": "RecurrentNet",
"shared": true,
"cell_type": "GRU",
"fc_hid_layers": [],
"hid_layers_activation": "relu",
"rnn_hidden_size": 32,
"rnn_num_layers": 1,
"seq_len": 4,
"clip_grad_val": 10.0,
"use_same_optim": false,
"actor_optim_spec": {
"name": "Adam",
"lr": 0.02
},
"critic_optim_spec": {
"name": "Adam",
"lr": 0.02
},
"lr_scheduler_spec": {
"name": "StepLR",
"step_size": 2000,
"gamma": 0.9,
},
"gpu": false
}
}],
"env": [{
"name": "Pendulum-v0",
"max_t": null,
"max_frame": 500,
}],
"body": {
"product": "outer",
"num": 1
},
"meta": {
"distributed": false,
"eval_frequency": 1000,
"max_session": 4,
"max_trial": 100,
"search": "RandomSearch"
},
"search": {
"agent": [{
"net": {
"hid_layers__choice": [
[32],
[64],
[32, 16],
[64, 32]
],
"hid_layers_activation__choice": ["sigmoid", "relu", "tanh"],
"actor_optim_spec": {
"lr__uniform": [0.0001, 0.2]
},
"critic_optim_spec": {
"lr__uniform": [0.0001, 0.2]
}
}
}]
}
},
"ppo_rnn_separate_pendulum": {
"agent": [{
"name": "PPO",
"algorithm": {
"name": "PPO",
"action_pdtype": "default",
"action_policy": "default",
"explore_var_spec": null,
"gamma": 0.99,
"lam": 0.95,
"clip_eps_spec": {
"name": "linear_decay",
"start_val": 0.10,
"end_val": 0.01,
"start_step": 1000,
"end_step": 5000,
},
"entropy_coef_spec": {
"name": "linear_decay",
"start_val": 0.01,
"end_val": 0.001,
"start_step": 1000,
"end_step": 5000,
},
"val_loss_coef": 0.1,
"time_horizon": 8,
"training_epoch": 8
},
"memory": {
"name": "OnPolicyBatchReplay"
},
"net": {
"type": "RecurrentNet",
"shared": false,
"cell_type": "GRU",
"fc_hid_layers": [],
"hid_layers_activation": "relu",
"rnn_hidden_size": 32,
"rnn_num_layers": 1,
"seq_len": 4,
"clip_grad_val": 10.0,
"use_same_optim": false,
"actor_optim_spec": {
"name": "Adam",
"lr": 0.02
},
"critic_optim_spec": {
"name": "Adam",
"lr": 0.02
},
"lr_scheduler_spec": {
"name": "StepLR",
"step_size": 2000,
"gamma": 0.9,
},
"gpu": false
}
}],
"env": [{
"name": "Pendulum-v0",
"max_t": null,
"max_frame": 500,
}],
"body": {
"product": "outer",
"num": 1
},
"meta": {
"distributed": false,
"eval_frequency": 1000,
"max_session": 4,
"max_trial": 100,
"search": "RandomSearch"
},
"search": {
"agent": [{
"net": {
"hid_layers__choice": [
[32],
[64],
[32, 16],
[64, 32]
],
"hid_layers_activation__choice": ["sigmoid", "relu", "tanh"],
"actor_optim_spec": {
"lr__uniform": [0.0001, 0.2]
},
"critic_optim_spec": {
"lr__uniform": [0.0001, 0.2]
}
}
}]
}
},
}