diff --git a/GPy/models/state_space.py b/GPy/models/state_space.py
index a3055038..1d202fa2 100644
--- a/GPy/models/state_space.py
+++ b/GPy/models/state_space.py
@@ -19,10 +19,19 @@ class StateSpace(Model):
         else:
             self.kern = kernel
 
+        #TODO:assert something about the kernel being an AR kernel?
+
+
     def set_params(self, x):
         self.kern.set_params(x[:self.kern.num_params_transformed()])
         self.sigma2 = x[-1]
 
+        #get the new model matrices from the kernel
+
+        #run the kalman filter
+
+        #run the rts smoother
+
     def get_params(self):
         return np.append(self.kern.get_params_transformed(), self.sigma2)
 
@@ -32,14 +41,34 @@ class StateSpace(Model):
     def log_likelihood(self):
         #TODO
 
-    def log_likelihood_gradients(self):
+    def _log_likelihood_gradients(self):
         #TODO
+        dL_dsigma2 = ???
+        dL_dtheta = self.kern.dL_dtheta_via_FL(self.dL_dF, self.dL_dL)
+        return np.hstack((dL_dtheta, dL_dsigma2))
 
-    def predict_raw(self):
+    def predict_raw(self, Xnew):
         #TODO
+        #make a single matrix containing traingin and testing points
+
+        #sort the matrix (save the order
+
+        #run the kalman filter again
+
+        #run the smoother
+
+        #put the data back in the original order, return the posterior of the state
 
     def predict(self):
         #TODO
 
+        #run the kalman filter to get the state, add the noise variance to the state variance
+
     def plot(self):
         #TODO
+
+    def posterior_samples_f(self,X,size=10):
+        #TODO
+
+    def posterior_samples(self, X, size=10):
+        #TODO