def test_ges_vs_lr(loop_num=1, n=30, p=7, c=4):
# update_pri_params()
lr_loss_list = []
ges_loss_list = []
count = 0
for i in tqdm(range(loop_num)):
lr_ate = None
ges_ate = None
while ges_ate is None or lr_ate is None:
count += 1
dag_edge = random_dag(p, c)
pri_dict = set_pri_params(dag_edge)
params = generate_params(pri_dict)
x_do = 5.0
print("------Intervention------\nX = {}".format(str(x_do)))
true_ate = calculate_true_ate(dag_edge, params, x_do)
df = set_dataframe(n, dag_edge, params, write_bool=False)
lr_ate = simple_linear_reg(df=df)
ges_ate = ges_method(df=df)
lr_loss = squared_loss(predicted_ate=lr_ate, true_ate=true_ate)
ges_loss = squared_loss(predicted_ate=ges_ate, true_ate=true_ate)
lr_loss_list.append(lr_loss)
ges_loss_list.append(ges_loss)
lr_res = np.mean(lr_loss_list)
ges_res = np.mean(ges_loss_list)
print("*****MSE Of LinearReg*****\n{}".format(lr_res))
print("*****MSE Of GES*****\n{}".format(ges_res))
print(count)
return [lr_res, ges_res]
def test_rep_kind(loop_num=100, n=10, p=5, c=4, rep_num=10):
logger.debug("loop_num={}, n={}, p={}, c={}, rep_num={}".format(loop_num, n, p, c, rep_num))
lr_loss_list = []
ges_loss_list = []
bayes_loss_list = {}
for rep_kind in [1, 2, 3]:
bayes_loss_list[rep_kind] = []
for i in tqdm(range(loop_num)):
lr_ate = None
ges_ate = None
bayes_ate = {}
for rep_kind in [1, 2, 3]:
bayes_ate[rep_kind] = None
while all([v is None for v in bayes_ate.values()]) or ges_ate is None or lr_ate is None:
dag_edge = random_dag(p, c)
pri_dict = set_pri_params(dag_edge)
params = generate_params(pri_dict)
x_do = 5.0
print("------Intervention------\nX = {}".format(str(x_do)))
true_ate = calculate_true_ate(dag_edge, params, x_do)
df = set_dataframe(n, dag_edge, params, write_bool=False)
lr_ate = simple_linear_reg(df=df)
ges_ate = ges_method(df=df)
for rep_kind in tqdm([1, 2, 3]):
bayes_ate[rep_kind] = bayes_method(df=df, c=c, rep_num=rep_num, rep_kind=rep_kind)
lr_loss_list.append(squared_loss(true_ate=true_ate, predicted_ate=lr_ate))
ges_loss_list.append(squared_loss(true_ate=true_ate, predicted_ate=ges_ate))
for rep_kind in [1, 2, 3]:
bayes_loss_list[rep_kind].append(squared_loss(true_ate=true_ate, predicted_ate=bayes_ate[rep_kind]))
row = [np.mean(lr_loss_list), np.mean(ges_loss_list)]
colomn = ['MSE of LinearReg', 'MSE of GES']
for rep_kind in [1, 2, 3]:
row.append(np.mean(bayes_loss_list[rep_kind]))
colomn.append('MSE of Bayes on rep_kind={}'.format(rep_kind))
df_res = pd.DataFrame([row], columns=colomn)
# Write to csv file
dt_now = datetime.datetime.now()
sheet = dt_now.strftime('%Y-%m-%d-%H-%M-%S')
path = "./data_files/loss_result_rep_kind.xlsx"
if os.path.exists(path):
with pd.ExcelWriter(path, engine="openpyxl", mode="a") as writer:
df_res.to_excel(writer, sheet_name=sheet, index=False)
else:
with pd.ExcelWriter(path, engine="openpyxl") as writer:
df_res.to_excel(writer, sheet_name=sheet, index=False)
print("DONE!!\nWrite in the sheet={} of data_files/loss_result_rep_kind.xlsx".format(sheet))
def test(loop_num=5, n=10, p=7, c=4, rep_num=4, rep_kind=1, update_param=False, update_epsilon=False):
if update_param:
update_pri_params()
if update_epsilon:
update_epsilon_dict()
lr_loss_list = []
ges_loss_list = []
bayes_loss_list = []
for i in tqdm(range(loop_num)):
lr_ate = None
bayes_ate = None
ges_ate = None
while bayes_ate is None or ges_ate is None or lr_ate is None:
dag_edge = random_dag(p, c)
pri_dict = set_pri_params(dag_edge)
params = generate_params(pri_dict)
x_do = 5.0
print("------Intervention------\nX = {}".format(str(x_do)))
true_ate = calculate_true_ate(dag_edge, params, x_do)
df = set_dataframe(n, dag_edge, params, write_bool=False)
lr_ate = simple_linear_reg(df=df)
ges_ate = ges_method(df=df)
if ges_ate is None:
continue
bayes_ate = bayes_method(df=df, c=c, rep_num=rep_num, rep_kind=rep_kind)
lr_loss = squared_loss(predicted_ate=lr_ate, true_ate=true_ate)
ges_loss = squared_loss(predicted_ate=ges_ate, true_ate=true_ate)
bayes_loss = squared_loss(true_ate=true_ate, predicted_ate=bayes_ate)
# if lr_loss + 50 < bayes_loss:
# continue
lr_loss_list.append(lr_loss)
ges_loss_list.append(ges_loss)
bayes_loss_list.append(bayes_loss)
print("******MSE******\n{}, {}, {}".format(np.mean(lr_loss_list), np.mean(ges_loss_list),
np.mean(bayes_loss_list)))
lr_res = np.mean(lr_loss_list)
lr_std = np.std(lr_loss_list)
ges_res = np.mean(ges_loss_list)
ges_std = np.std(ges_loss_list)
bayes_res = np.mean(bayes_loss_list)
bayes_std = np.std(bayes_loss_list)
print("*****MSE Of LinearReg*****\n{}".format(lr_res))
print("*****MSE Of GES*****\n{}".format(ges_res))
print("*****MSE Of Bayes*****\n{}".format(bayes_res))
logger.debug("loop_num={}, n={}, p={}, c={}, rep_num={}, rep_kind={}".format(loop_num, n, p, c, rep_num, rep_kind))
logger.debug("MSE Of LinearReg={}".format(lr_res))
logger.debug("STD of LinearReg={}".format(lr_std))
logger.debug("MSE Of GES={}".format(ges_res))
logger.debug("STD of GES={}".format(ges_std))
logger.debug("MSE Of Bayes={}".format(bayes_res))
logger.debug("STD of Bayes={}".format(bayes_std))
rows = []
for lr, ges, bayes in zip(lr_loss_list, ges_loss_list, bayes_loss_list):
rows.append([lr, ges, bayes])
write_excel(rows=rows, col_name=['lr', 'GES', 'Bayes'], n=n)
return [lr_res, ges_res, bayes_res, lr_std, ges_std, bayes_std]