def hdp_sample_only():
print('hdp_sample only')
round = 800
rand_seed = np.random.randint(0, 10000000)
res_hdp = run_hdp(rounds=round, seed=rand_seed)
eval_res = OneRoundEvaluation(res_list=[res_hdp])
eval_res.plot_all()
def compare_hdp_hdpsample():
print('hdp and sample')
round = 1000
rand_seed = np.random.randint(0, 10000000)
res_hdp = run_hdp(rounds=round, seed=rand_seed)
res_hdp_sample = run_hdp_sample(rounds=round, seed=rand_seed)
eval_res = OneRoundEvaluation(res_list=[res_hdp, res_hdp_sample])
eval_res.plot_all()
def vi_test():
exp_round = 3
res_list = []
for t in range(exp_round):
round = 800
rand_seed = np.random.randint(0, 10000000)
res_list.append(
run_vi(rounds=round, seed=rand_seed, name='VI_' + str(t + 1)), )
eval_res = OneRoundEvaluation(res_list=res_list)
eval_res.plot_all()
def hdp_five_times():
print('hdp 5')
res_list = []
for t in range(5):
round = 800
rand_seed = np.random.randint(0, 10000000)
res_list.append(
run_hdp(rounds=round, seed=rand_seed, name='hdp_' + str(t + 1)), )
eval_res = OneRoundEvaluation(res_list=res_list)
eval_res.plot_all()
def run_adhdp_only():
round = 3200
predict_round = 800
res_list = []
rand_seed = np.random.randint(0, 10000000)
# rand_seed = 320743
res_list.append(
run_adhdp(rounds=round,
seed=rand_seed,
name='ADHDP',
predict_round=predict_round))
eval_res = OneRoundEvaluation(res_list=res_list)
eval_res.plot_all()
def ADHDP_Off_Policy_only():
round = 3200
res_list = []
rand_seed = np.random.randint(0, 10000000)
# rand_seed = 320743
res_list.append(
run_adhdp_offpolicy(rounds=round,
seed=rand_seed,
name='ADHDP',
train_rounds=5,
train_step_in_round=200))
eval_res = OneRoundEvaluation(res_list=res_list)
eval_res.plot_all()
def vi_compare_sample():
exp_round = 1
res_list = []
predict_round = 800
rand_seed = np.random.randint(0, 10000000)
for t in range(exp_round):
round = 800
res_list.append(
run_vi(rounds=round,
seed=rand_seed,
name='VI',
predict_round=predict_round), )
for t in range(exp_round):
round = 800
res_list.append(
run_vi_sample(rounds=round,
seed=rand_seed,
name='VI_sample',
predict_round=predict_round), )
eval_res = OneRoundEvaluation(res_list=res_list)
mse_dict = eval_res.plot_all()
mse_vi.append(mse_dict['VI'])
mse_vi_sample.append(mse_dict['VI_sample'])
def vi_compare_hdp():
round = 800
predict_round = 800
res_list = []
rand_seed = np.random.randint(0, 10000000)
# rand_seed = 320743
res_list.append(
run_vi(rounds=round,
seed=rand_seed,
name='VI',
predict_round=predict_round))
res_list.append(
run_hdp(rounds=round,
seed=rand_seed,
name='HDP',
predict_round=predict_round))
eval_res = OneRoundEvaluation(res_list=res_list)
eval_res.plot_all()
def vi_diff_capacity(capacity_list=None):
if capacity_list is None:
capacity = range(1, 12, 3)
predict_round = 800
res_list = []
rand_seed = np.random.randint(0, 10000000)
rand_seed = 3309316
for capacity in capacity_list:
round = 800
res_list.append(
run_vi(rounds=round,
seed=rand_seed,
capacity=capacity,
name='Replay: ' + str(capacity),
predict_round=predict_round))
eval_res = OneRoundEvaluation(res_list=res_list)
eval_res.plot_all()
def run_adhdp(rounds=1000, seed=random.randint(0, 1000000)):
print('seed :', seed)
random.seed(seed)
np.random.seed(seed)
penalty = Quadratic(**penalty_para)
env_adhdp = Thickener(
penalty_calculator=penalty,
**thickner_para,
random_seed=seed,
)
env_adhdp.reset()
res1 = OneRoundExp(controller=adhdp,
env=env_adhdp,
max_step=rounds,
exp_name='ADHDP').run()
eval_res = OneRoundEvaluation(res_list=[res1])
eval_res.plot_all()
def vi_optim_compare():
exp_round = 1
res_list = []
predict_round = 800
rand_seed = np.random.randint(0, 10000000)
opt_list = ['adagrad', 'RMSprop', 'adam', 'sgd']
opt_list = ['adam', 'sgd']
for opt_name in opt_list:
round = 800
res_list.append(
run_vi(rounds=round,
seed=rand_seed,
name='VI_' + opt_name,
predict_round=predict_round,
u_optim=opt_name), )
eval_res = OneRoundEvaluation(res_list=res_list)
mse_dict = eval_res.plot_all()