def initialize_modelM(self):
nn = Ort_NN(dims=[self.Xprobit.shape[1], 20, self.proj_dim],
N=0,
proj_dim=0,
name=None)
k_list, gp_nnjoint = bloc_diag_initialize_models(
x=np.copy(self.Xprobit),
y=np.copy(self.Ynorm),
input_dim=self.proj_dim,
model='diagonal_joint',
kernel='Matern52',
ARD=True,
nn=nn,
decomp=self.decomposition) # last kernel is the Y kernel
# k2_list, gp_nnjoint_test = initialize_m_models(x=np.copy(self.Xprobit), y=np.copy(self.Ynorm),
# input_dim=self.proj_dim,
# model='joint',
# kernel='Matern52',
# ARD=True,
# nn=nn,
# decomp=self.decomposition) # last kernel is the Y kernel
gp_nnjoint.likelihood.variance = 1e-06 # 0.001
# gp_nnjoint_test.likelihood.variance = 1e-06
return k_list, gp_nnjoint, nn
def initialize_modelM(self):
nn = Ort_NN(dims=[self.Xprobit.shape[1], 20, self.proj_dim],
N=0,
proj_dim=0,
name=None)
# nn = NN(dims=[self.Xprobit.shape[1], 20, self.proj_dim], N=0, proj_dim=0,
# name=None)
k_list, gp_nnjoint = bloc_diag_initialize_models(
x=np.copy(self.Xprobit),
y=np.copy(self.Ynorm),
input_dim=self.proj_dim,
model='joint',
kernel='Matern52',
ARD=True,
nn=nn,
decomp=self.decomposition) # last kernel is the Y kernel
# k_list, gp_nnjoint = initialize_m_models(x=np.copy(self.Xprobit), y=np.copy(self.Ynorm),
# input_dim=self.proj_dim,
# model='joint',
# kernel='Matern52',
# ARD=True,
# nn=nn,
# decomp=self.decomposition) # last kernel is the Y kernel
# kern_joint = Kstack(k_list)
# gp_nnjoint = NN_MoGPR(X=np.copy(self.Xprobit), Y=np.copy(self.Ynorm), kern=kern_joint, nn=nn, Mo_dim=self.Mo_dim)
gp_nnjoint.likelihood.variance = 1e-06 # 0.001
return k_list, gp_nnjoint, nn
def initialize_modelM(self):
nn = Ort_NN(dims=[self.data_x.shape[1], 20, self.proj_dim], N=0, proj_dim=0,
name=None)
k_list, gp_nnjoint = initialize_m_models(x=np.copy(self.Xnorm), y=np.copy(self.Ynorm),
# k_list, gp_nnjoint = initialize_m_models(x=np.copy(self.X_inf), y=np.copy(self.Ynorm),
input_dim=self.proj_dim,
model='BLR',
kernel='Matern52',
ARD=True,
nn=nn) # last kernel is the Manifold GP kernel
gp_nnjoint.likelihood.variance = 1e-06 # 0.001
return k_list, gp_nnjoint, nn
def initialize_modelM(self):
nn = Ort_NN(dims=[self.Xnorm.shape[1], 20, self.proj_dim], N=0, proj_dim=0,
name=None)
kernm, gpm = initialize_m_models(x=np.copy(self.Xnorm), y=np.copy(self.Ynorm),
input_dim=self.proj_dim,
model='encoder',
kernel='Matern52',
ARD=True,
nn=nn,
decomp=None)
gpm.likelihood.variance = 0.01
return kernm, gpm
def initialize_modelM(self):
nn = Ort_NN(dims=[self.Xnorm.shape[1], 20, self.proj_dim],
N=0,
proj_dim=0,
name=None)
# nn = NN(dims=[self.Xnorm.shape[1], 20, self.proj_dim], N=0, proj_dim=0,
# name=None)
kernm, gpm = initialize_m_models(x=np.copy(self.Xnorm),
y=np.copy(self.Ynorm),
input_dim=self.proj_dim,
model='encoder',
kernel='Matern52',
ARD=True,
nn=nn,
decomp=None)
gpm.likelihood.variance = 0.01
# k1 = gpflow.kernels.RBF(input_dim=self.proj_dim, ARD=True, active_dims=list(range(self.proj_dim)))
# coreg = gpflow.kernels.Coregion(input_dim=1, output_dim=self.Xnorm.shape[1] + 1,
# rank=self.Xnorm.shape[1] + 1, active_dims=[self.proj_dim])
# coreg.W = np.random.randn(self.Xnorm.shape[1] + 1, self.Xnorm.shape[1] + 1)
# values = np.random.normal(loc=0., scale=1., size=[len(self.binary_tree_list)])
# tree_coregion = TreeCoregion(input_dim=1, output_dim=self.Xnorm.shape[1] + 1,
# indices_tree=self.binary_tree_list, values=values,
# active_dims=[self.proj_dim])
# kc = k1 * coreg
# kc = k1 * tree_coregion
# nn = NN(dims=[self.Xnorm.shape[1], 20, self.proj_dim], N=0, proj_dim=0,
# name=None) # otherwise re-initialized at each BO iteration
# nn = NN(dims=[self.Xnorm.shape[1], 6, self.proj_dim], N=0, proj_dim=0,
# name=None) # otherwise re-initialized at each BO iteration
# nn_mogp = NN_MOGPR(X=np.copy(self.Xnorm), Y=np.copy(self.Ynorm), kern=kc, nn=nn)
# nn_mogp.likelihood.variance = 0.0001
return kernm, gpm, nn