pseudo EM algorithm for EP and maybe Laplace.

GPy/likelihoods/EP.py

@@ -33,6 +33,15 @@ class EP(likelihood):
         self.Z = 0
         self.YYT = None
 
+    def restart(self):
+        self.tau_tilde = np.zeros(self.N)
+        self.v_tilde = np.zeros(self.N)
+        self.Y = np.zeros((self.N,1))
+        self.covariance_matrix = np.eye(self.N)
+        self.precision = np.ones(self.N)[:,None]
+        self.Z = 0
+        self.YYT = None
+
     def predictive_values(self,mu,var,full_cov):
         if full_cov:
             raise NotImplementedError, "Cannot make correlated predictions with an EP likelihood"

GPy/likelihoods/likelihood.py

@@ -1,4 +1,5 @@
 import numpy as np
+import copy
 
 class likelihood:
     """
@@ -37,4 +38,6 @@ class likelihood:
     def predictive_values(self, mu, var):
         raise NotImplementedError
 
-
+    def copy(self):
+        """ Returns a (deep) copy of the current likelihood """
+        return copy.deepcopy(self)