def run_hdp(
rounds=1000,
seed=random.randint(0, 1000000),
name='VI',
capacity=2,
batch_size=2,
predict_round=3000,
u_optim='adam',
):
print('seed :', seed)
torch.manual_seed(seed)
np.random.seed(seed)
random.seed(seed)
vi = new_hdp(capacity=capacity, batch_size=batch_size)
penalty = Quadratic(**penalty_para)
env_vi = Thickener(
penalty_calculator=penalty,
**thickner_para,
)
res1 = OneRoundExp(controller=vi,
env=env_vi,
max_step=rounds,
exp_name=name).run()
return res1
def run_dhp(
rounds=800,
seed=random.randint(0, 1000000),
name='DHP',
capacity=2,
predict_round=3000,
u_optim='adam',
):
#seed = 8312279
print('seed :', seed)
torch.manual_seed(seed)
random.seed(seed)
np.random.seed(seed)
dhp = new_dhp()
penalty = Quadratic(**penalty_para)
env_dhp = Thickener(
penalty_calculator=penalty,
**thickner_para,
)
res1 = OneRoundExp(controller=dhp,
env=env_dhp,
max_step=rounds,
exp_name=name).run()
return res1
def run_vi(
rounds=1000,
seed=random.randint(0, 1000000),
name='VI',
capacity=2,
predict_round=3000,
u_optim='adam',
):
print('seed :', seed)
torch.manual_seed(seed)
random.seed(seed)
np.random.seed(seed)
vi_para = {'gamma': 0.2}
vi = new_vi()
penalty = Quadratic(**penalty_para)
env_vi = Thickener(
penalty_calculator=penalty,
**thickner_para,
)
res1 = OneRoundExp(controller=vi,
env=env_vi,
max_step=rounds,
exp_name=name).run()
print(name, ':', vi.u_iter_times * 1.0 / rounds)
return res1
def run_ILPL(rounds=1000,
seed=random.randint(0, 1000000),
name='ILPL',
predict_round=800):
print('seed :', seed)
torch.manual_seed(seed)
random.seed(seed)
np.random.seed(seed)
ilpl = new_ILPL()
penalty = Quadratic(**penalty_para)
env_ILPL = Thickener(
penalty_calculator=penalty,
**thickner_para,
)
res1 = OneRoundExp(controller=ilpl,
env=env_ILPL,
max_step=rounds,
exp_name=name).run()
return res1
def run_hdp(rounds=1000,
seed=random.randint(0, 1000000),
name='HDP',
predict_round=800):
print('seed :', seed)
hdp_para = {'gamma': 0.2}
torch.manual_seed(seed)
hdp = new_hdp()
penalty = Quadratic(**penalty_para)
env_hdp = Thickener(
penalty_calculator=penalty,
**thickner_para,
)
res1 = OneRoundExp(controller=hdp,
env=env_hdp,
max_step=rounds,
exp_name=name).run()
return res1