Convert SSGP updates to use assign method

sgptools/models/core/osgpr.py

@@ -67,18 +67,16 @@ def update(self, data):
         self.X, self.Y = self.data = gpflow.models.util.data_input_to_tensor(data)
         self.num_data = self.X.shape[0]
 
-        self.Z_old = tf.Variable(self.inducing_variable.Z.numpy(), 
-                                 shape=tf.TensorShape(None), 
-                                 trainable=False)
+        self.Z_old.assign(self.inducing_variable.Z.numpy())
 
         # Get posterior mean and covariance for the old inducing points
         mu_old, Su_old = self.predict_f(self.Z_old, full_cov=True)
-        self.mu_old = tf.Variable(mu_old, shape=tf.TensorShape(None), trainable=False)
-        self.Su_old = tf.Variable(Su_old, shape=tf.TensorShape(None), trainable=False)
-        
+        self.mu_old.assign(mu_old.numpy())
+        self.Su_old.assign(Su_old.numpy())
+
         # Get the prior covariance matrix for the old inducing points
         Kaa_old = self.kernel(self.Z_old)
-        self.Kaa_old = tf.Variable(Kaa_old, shape=tf.TensorShape(None), trainable=False)
+        self.Kaa_old.assign(Kaa_old.numpy())
 
     def _common_terms(self):
         Mb = self.inducing_variable.num_inducing