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)
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_dhp_vi(
rounds=1000,
seed=random.randint(0, 1000000),
name='DHPVI',
capacity=2,
predict_round=3000,
u_optim='adam',
):
print('seed :', seed)
torch.manual_seed(seed)
dhp_vi_para = {
#'gamma': 0.2
}
dhp_vi = new_dhp_vi()
specific_penalty_para = copy.deepcopy(penalty_para)
specific_penalty_para['S'] = [0.0001, 0.0008]
penalty = Quadratic(**specific_penalty_para)
env_dhp_vi = Thickener(
penalty_calculator=penalty,
**thickner_para,
)
res1 = OneRoundExp(controller=dhp_vi,
env=env_dhp_vi,
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
}
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
def run_dhp(rounds=800,seed=random.randint(0,1000000),name='DHP',capacity=2,
predict_round=3000,u_optim='adam',):
print('seed :',seed)
torch.manual_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_adhdp(rounds=1000,
seed=random.randint(0, 1000000),
name='ADHDP',
predict_round=800,
random_act=False):
print('seed :', seed)
torch.manual_seed(seed)
random.seed(seed)
np.random.seed(seed)
adhdp = new_adhdp(random_act=random_act)
penalty = Quadratic(**penalty_para)
env_hdp = Thickener(
penalty_calculator=penalty,
random_seed=seed,
**thickner_para,
)
res1 = OneRoundExp(controller=adhdp,
env=env_hdp,
max_step=rounds,
exp_name=name).run()
return res1