def new_vi(capacity=2, batch_size=2):
capacity = capacity
predict_round=3000
u_optim='adam'
gamma=0.6
replay_vi = ReplayBuffer(capacity=capacity)
env_VI = Thickener(
noise_p=0.03,
noise_in=True,
)
exploration = No_Exploration()
print('make new vi controller')
vi = VI(
replay_buffer = replay_vi,
u_bounds = env_VI.u_bounds,
#exploration = None,
exploration = exploration,
env=env_VI,
predict_training_rounds=predict_round,
gamma=gamma,
batch_size = batch_size,
predict_batch_size=32,
model_nn_error_limit = 0.0008,
critic_nn_error_limit = 0.001,
actor_nn_error_limit = 0.001,
actor_nn_lr = 0.005,
critic_nn_lr = 0.01,
model_nn_lr = 0.01,
indice_y = None,
indice_y_star = None,
indice_c=None,
hidden_model = 10,
hidden_critic = 14,
hidden_actor = 14,
predict_epoch= 30,
Nc=1000,
u_optim=u_optim,
img_path=EXP_NAME
)
env_VI.reset()
vi.train_identification_model()
#vi.test_predict_model(test_rounds=100)
return vi
def new_dhp():
capacity = 3
predict_round = 3000
gamma = 0.6
replay_DHP = ReplayBuffer(capacity=capacity)
env_DHP = Thickener(
noise_p=0.03,
noise_in=True,
)
exploration = No_Exploration()
print('make new dhp controller')
dhp = DHP(
replay_buffer=replay_DHP,
u_bounds=env_DHP.u_bounds,
#exploration = None,
exploration=exploration,
env=env_DHP,
predict_training_rounds=predict_round,
gamma=gamma,
batch_size=2,
predict_batch_size=32,
model_nn_error_limit=0.0008,
critic_nn_error_limit=0.001,
actor_nn_error_limit=0.0001,
# 0.005
actor_nn_lr=0.008,
critic_nn_lr=0.01,
model_nn_lr=0.01,
indice_y=None,
indice_y_star=None,
indice_c=None,
hidden_model=10,
hidden_critic=12,
hidden_actor=14,
predict_epoch=30,
Na=200,
Nc=300,
img_path=EXP_NAME)
env_DHP.reset()
dhp.train_identification_model()
return dhp
def new_ILPL():
predict_round = 3000
gamma = 0.6
replay_ILPL = ReplayBuffer(capacity=4)
env_ILPL = Thickener(
noise_p=0.03,
noise_in=True,
)
exploration = No_Exploration()
print('make new ilpl controller')
ilpl = ILPL(
replay_buffer=replay_ILPL,
u_bounds=env_ILPL.u_bounds,
#exploration = None,
exploration=exploration,
env=env_ILPL,
predict_training_rounds=predict_round,
gamma=gamma,
batch_size=2,
predict_batch_size=32,
model_nn_error_limit=0.0008,
critic_nn_error_limit=0.1,
actor_nn_error_limit=0.001,
# 0.005
actor_nn_lr=0.001,
critic_nn_lr=0.02,
model_nn_lr=0.01,
indice_y=None,
indice_y_star=None,
indice_c=None,
hidden_model=10,
hidden_critic=14,
hidden_actor=14,
predict_epoch=30,
Na=220,
Nc=500,
img_path=EXP_NAME)
env_ILPL.reset()
ilpl.train_identification_model()
return ilpl