# test_abnormal = data_abnormal.drop(train_abnormal.index)
# x, y = train.iloc[:, :-1].values, train.iloc[:, -1].values.reshape(-1,1)
x, y = x_train.iloc[:, :].values, y_train.iloc[:, :].values.reshape(-1,1)
opts = {
'min_samples': 0,
'X': x,
'qd': 4,
'max_depth': 10,
'max_samples': 6000,
'log': True,
'jump': True,
'reduce_branching': True
}
root_region, gps_ = build_bins(**opts)
#root_region, gps_ = build(X=x, delta_divisor=3, max_depth=2)
root, gps, gps1, gps2 = structure(root_region, gp_types=['rbf']) #modified
for i, gp in enumerate(gps):
idx = query(x, gp.mins, gp.maxs)
gp.x, gp.y = x[idx], y[idx]
print(f"Training GP set1 {i+1}/{len(gps)} ({len(idx)})" #modified
gp.init(cuda=True)
##modified
for i, gp in enumerate(gps1):
idx = query(x, gp.mins, gp.maxs)
gp.x, gp.y = x[idx], y[idx]
test = data.drop(train.index)
x, y = train.iloc[:, :-1].values, train.iloc[:, -1].values.reshape(-1, 1)
opts = {
'min_samples': 0,
'X': x,
'qd': 3,
'max_depth': 2,
'max_samples': 10**10,
'log': True,
'min_samples': 0,
'jump': True,
'reduce_branching': True
}
root_region, gps_ = build_bins(**opts)
root, gps = structure(root_region, gp_types=['rbf'])
for i, gp in enumerate(gps):
idx = query(x, gp.mins, gp.maxs)
gp.x, gp.y = x[idx], y[idx]
print(f"Training GP {i+1}/{len(gps)} ({len(idx)})")
gp.init(cuda=True)
root.update()
for smudge in np.arange(0, 0.5, 0.05):
mu_s, cov_s = root.forward(test.iloc[:, 0:-1].values, smudge=smudge)
mu_s = (mu_s.ravel() * dstd.iloc[-1]) + dmean.iloc[-1]
mu_t = (test.iloc[:, -1] * dstd.iloc[-1]) + dmean.iloc[-1]
sqe = (mu_s - mu_t.values)**2
rmse = np.sqrt(sqe.sum() / len(test))
for kkk in range(rerun):
opts = {
'min_samples': 0,
'X': x,
'qd': 9,
'max_depth': 100,
'max_samples': 5100,
'log': True,
'jump': True,
'reduce_branching': True
}
root_region, gps_ = build_bins(**opts)
root, gps = structure(root_region, gp_types=['matern1.5_ard']) #modified
rmse = 0
mae = 0
nlpd_all = []
count = 0
mid_LMM = np.zeros((y_d, epoch))
for k in range(y_d):
y_loop = y[:, k].reshape(-1, 1)
# for i, gp in enumerate(gps):
# idx = query(x, gp.mins, gp.maxs)
# gp.x, gp.y = x[idx], y_loop[idx]
# print(f"Training GP set1 {i+1}/{len(gps)} ({len(idx)})") #modified
# gp.init(cuda=True)
# lr = 0.1
# steps = 200