Merge branch 'devel' of github.com:SheffieldML/GPy into devel

GPy/core/model.py

@@ -264,8 +264,11 @@ class model(parameterised):
         return - LL_gradients - prior_gradients
 
     def objective_and_gradients(self, x):
-        obj_f = self.objective_function(x)
-        obj_grads = self.objective_function_gradients(x)
+        self._set_params_transformed(x)
+        obj_f =  -self.log_likelihood() - self.log_prior()
+        LL_gradients = self._transform_gradients(self._log_likelihood_gradients())
+        prior_gradients = self._transform_gradients(self._log_prior_gradients())
+        obj_grads = - LL_gradients - prior_gradients
         return obj_f, obj_grads
 
     def optimize(self, optimizer=None, start=None, **kwargs):

GPy/inference/SGD.py

@@ -220,7 +220,7 @@ class opt_SGD(Optimizer):
             b = len(features)/self.batch_size
             features = [features[i::b] for i in range(b)]
             NLL = []
-
+            import pylab as plt
             for count, j in enumerate(features):
                 self.model.D = len(j)
                 self.model.likelihood.D = len(j)
@@ -230,6 +230,8 @@ class opt_SGD(Optimizer):
                     shapes = self.get_param_shapes(N, Q)
                     f, step, Nj = self.step_with_missing_data(f_fp, X, step, shapes)
                 else:
+                    self.model.likelihood.YYT = np.dot(self.model.likelihood.Y, self.model.likelihood.Y.T)
+                    self.model.likelihood.trYYT = np.trace(self.model.likelihood.YYT)
                     Nj = N
                     f, fp = f_fp(self.x_opt)
                     step = self.momentum * step + self.learning_rate * fp
@@ -244,7 +246,11 @@ class opt_SGD(Optimizer):
                 NLL.append(f)
 
                 self.fopt_trace.append(f)
+                # fig = plt.figure('traces')
+                # plt.clf()
+                # plt.plot(self.param_traces['noise'])
 
+                # import pdb; pdb.set_trace()
                 # for k in self.param_traces.keys():
                 #     self.param_traces[k].append(self.model.get(k)[0])
 

GPy/models/Bayesian_GPLVM.py

@@ -69,8 +69,8 @@ class Bayesian_GPLVM(sparse_GP, GPLVM):
         return dKL_dmu, dKL_dS
 
     def dL_dmuS(self):
-        dL_dmu_psi0, dL_dS_psi0 = self.kern.dpsi1_dmuS(self.dL_dpsi1, self.Z, self.X, self.X_variance)
-        dL_dmu_psi1, dL_dS_psi1 = self.kern.dpsi0_dmuS(self.dL_dpsi0, self.Z, self.X, self.X_variance)
+        dL_dmu_psi0, dL_dS_psi0 = self.kern.dpsi0_dmuS(self.dL_dpsi0, self.Z, self.X, self.X_variance)
+        dL_dmu_psi1, dL_dS_psi1 = self.kern.dpsi1_dmuS(self.dL_dpsi1, self.Z, self.X, self.X_variance)
         dL_dmu_psi2, dL_dS_psi2 = self.kern.dpsi2_dmuS(self.dL_dpsi2, self.Z, self.X, self.X_variance)
         dL_dmu = dL_dmu_psi0 + dL_dmu_psi1 + dL_dmu_psi2
         dL_dS = dL_dS_psi0 + dL_dS_psi1 + dL_dS_psi2

GPy/util/linalg.py

@@ -97,7 +97,7 @@ def jitchol_old(A,maxtries=5):
 
         raise linalg.LinAlgError,"not positive definite, even with jitter."
 
-def pdinv(A):
+def pdinv(A, *args):
     """
     :param A: A DxD pd numpy array
 
@@ -110,7 +110,7 @@ def pdinv(A):
     :rval logdet: the log of the determinant of A
     :rtype logdet: float64
     """
-    L = jitchol(A)
+    L = jitchol(A, *args)
     logdet = 2.*np.sum(np.log(np.diag(L)))
     Li = chol_inv(L)
     Ai = linalg.lapack.flapack.dpotri(L)[0]