def _initialize_networks(self):
""" Initialize all model networks """
if self.TYPE_PX == 'Gaussian':
self.Pz_x = dgm._init_Gauss_net(self.Z_DIM,
self.NUM_HIDDEN,
self.X_DIM,
'Pz_x_',
bn=self.batchnorm)
elif self.TYPE_PX == 'Bernoulli':
self.Pz_x = dgm._init_Cat_net(self.Z_DIM,
self.NUM_HIDDEN,
self.X_DIM,
'Pz_x_',
bn=self.batchnorm)
self.Pzx_y = dgm._init_Cat_bnn(self.Z_DIM + self.X_DIM,
self.NUM_HIDDEN, self.NUM_CLASSES,
'Pzx_y', self.initVar)
self.Wtilde = self._init_Wtilde(self.Z_DIM + self.X_DIM,
self.NUM_HIDDEN,
self.NUM_CLASSES,
'W_tilde_',
bn=self.batchnorm)
self.Qxy_z = dgm._init_Gauss_net(self.X_DIM + self.NUM_CLASSES,
self.NUM_HIDDEN,
self.Z_DIM,
'Qxy_z',
bn=self.batchnorm)
self.Qx_y = dgm._init_Cat_net(self.X_DIM,
self.NUM_HIDDEN,
self.NUM_CLASSES,
'Qx_y',
bn=self.batchnorm)
def _initialize_networks(self):
""" Initialize all model networks """
if self.TYPE_PX == 'Gaussian':
self.Pza_x = dgm._init_Gauss_net(self.Z_DIM+self.A_DIM, self.NUM_HIDDEN, self.X_DIM, 'Pza_x_', self.batchnorm)
elif self.TYPE_PX == 'Bernoulli':
self.Pza_x = dgm._init_Cat_net(self.Z_DIM+self.A_DIM, self.NUM_HIDDEN, self.X_DIM, 'Pza_x_', self.batchnorm)
self.Pzax_y = dgm._init_Cat_net(self.Z_DIM+self.A_DIM+self.X_DIM, self.NUM_HIDDEN, self.NUM_CLASSES, 'Pzax_y_', self.batchnorm)
self.Pz_a = dgm._init_Gauss_net(self.Z_DIM, self.NUM_HIDDEN, self.A_DIM, 'Pz_a_', self.batchnorm)
self.Qxya_z = dgm._init_Gauss_net(self.X_DIM+self.A_DIM+self.NUM_CLASSES, self.NUM_HIDDEN, self.Z_DIM, 'Qxya_z_', self.batchnorm)
self.Qx_a = dgm._init_Gauss_net(self.X_DIM, self.NUM_HIDDEN, self.A_DIM, 'Qx_a_', self.batchnorm)
self.Qxa_y = dgm._init_Cat_net(self.X_DIM+self.A_DIM, self.NUM_HIDDEN, self.NUM_CLASSES, 'Qxa_y_', self.batchnorm)
def _initialize_networks(self):
""" Initialize all model networks """
if self.TYPE_PX == 'Gaussian':
self.Pzy_x = dgm._init_Gauss_net(self.Z_DIM + self.NUM_CLASSES,
self.NUM_HIDDEN, self.X_DIM,
'Pzy_x', self.batchnorm)
elif self.TYPE_PX == 'Bernoulli':
self.Pzy_x = dgm._init_Cat_net(self.Z_DIM + self.NUM_CLASSES,
self.NUM_HIDDEN, self.X_DIM,
'Pzy_x', self.batchnorm)
self.Py = tf.constant(
(1. / self.NUM_CLASSES) * np.ones(shape=(self.NUM_CLASSES, )),
dtype=tf.float32)
self.Qxy_z = dgm._init_Gauss_net(self.X_DIM + self.NUM_CLASSES,
self.NUM_HIDDEN, self.Z_DIM, 'Qxy_z',
self.batchnorm)
self.Qx_y = dgm._init_Cat_net(self.X_DIM, self.NUM_HIDDEN,
self.NUM_CLASSES, 'Qx_y', self.batchnorm)
def _initialize_networks(self):
if self.TYPE_PX=='Gaussian':
self.Pz_x = dgm._init_Gauss_net(self.Z_DIM, self.NUM_HIDDEN, self.X_DIM, 'Pz_x_', self.batchnorm)
elif self.TYPE_PX=='Bernoulli':
self.Pz_x = dgm._init_Cat_net(self.Z_DIM, self.NUM_HIDDEN, self.X_DIM, 'Pz_x_', self.batchnorm)
self.Qx_z = dgm._init_Gauss_net(self.X_DIM, self.NUM_HIDDEN, self.Z_DIM, 'Qx_z_', self.batchnorm)