demo changed, bgplvm still broken

GPy/examples/dimensionality_reduction.py

@@ -170,26 +170,30 @@ def bgplvm_simulation(burnin='scg', plot_sim=False, max_f_eval=12):
     from GPy import kern
     reload(mrd); reload(kern)
 
+
     Y = Ylist[1]
 
-    k = kern.linear(Q, ARD=True) + kern.bias(Q, .0001) + kern.white(Q, .1)
+    k = kern.linear(Q, ARD=True) + kern.white(Q, .00001)  # + kern.bias(Q)
     m = Bayesian_GPLVM(Y, Q, init="PCA", M=M, kernel=k)
-    m.set('noise', Y.var() / 100.)
+    # m.set('noise',)
 #     m.auto_scale_factor = True
 #     m.scale_factor = 1.
-
     m.ensure_default_constraints()
 
+
     if burnin:
         print "initializing beta"
         cstr = "noise"
-        m.unconstrain(cstr); m.constrain_fixed(cstr)
+        m.unconstrain(cstr); m.constrain_fixed(cstr, Y.var() / 100.)
         m.optimize(burnin, messages=1, max_f_eval=max_f_eval)
 
         print "releasing beta"
         cstr = "noise"
         m.unconstrain(cstr);  m.constrain_positive(cstr)
 
+    true_X = np.hstack((slist[1], slist[3], 0. * np.ones((N, Q - 2))))
+    m.set('X_\d', true_X)
+    m.constrain_fixed("X_\d")
 
 # #     cstr = 'variance'
 # #     m.unconstrain(cstr), m.constrain_bounded(cstr, 1e-10, 1.)

GPy/models/Bayesian_GPLVM.py

@@ -82,7 +82,7 @@ class Bayesian_GPLVM(sparse_GP, GPLVM):
             self._set_params(self.oldps[-1], save_old=False)
 
     def dKL_dmuS(self):
-        dKL_dS = (1. - (1. / self.X_variance)) * 0.5
+        dKL_dS = (1. - (1. / (self.X_variance))) * 0.5
         dKL_dmu = self.X
         return dKL_dmu, dKL_dS
 
@@ -101,13 +101,26 @@ class Bayesian_GPLVM(sparse_GP, GPLVM):
         return 0.5 * (var_mean + var_S) - 0.5 * self.Q * self.N
 
     def log_likelihood(self):
-        return sparse_GP.log_likelihood(self) - self.KL_divergence()
+        ll = sparse_GP.log_likelihood(self)
+        kl = self.KL_divergence()
+        return ll + kl
 
     def _log_likelihood_gradients(self):
         dKL_dmu, dKL_dS = self.dKL_dmuS()
         dL_dmu, dL_dS = self.dL_dmuS()
         # TODO: find way to make faster
-        dbound_dmuS = np.hstack(((dL_dmu - dKL_dmu).flatten(), (dL_dS - dKL_dS).flatten()))
+
+        d_dmu = (dL_dmu + dKL_dmu).flatten()
+        d_dS = (dL_dS + dKL_dS).flatten()
+        # TEST KL: ====================
+        # d_dmu = (dKL_dmu).flatten()
+        # d_dS = (dKL_dS).flatten()
+        # ========================
+        # TEST L: ====================
+#         d_dmu = (dL_dmu).flatten()
+#         d_dS = (dL_dS).flatten()
+        # ========================
+        dbound_dmuS = np.hstack((d_dmu, d_dS))
         return np.hstack((dbound_dmuS.flatten(), sparse_GP._log_likelihood_gradients(self)))
 
     def plot_latent(self, which_indices=None, *args, **kwargs):

GPy/models/GP.py

@@ -261,6 +261,11 @@ class GP(model):
             m, var, lower, upper = self.predict(Xnew, slices=which_functions)
             gpplot(Xnew, m, lower, upper)
             pb.plot(Xu[which_data], self.likelihood.data[which_data], 'kx', mew=1.5)
+            if self.has_uncertain_inputs:
+                pb.errorbar(Xu[which_data, 0], self.likelihood.data[which_data, 0],
+                            xerr=2 * np.sqrt(self.X_variance[which_data, 0]),
+                            ecolor='k', fmt=None, elinewidth=.5, alpha=.5)
+
             ymin, ymax = min(np.append(self.likelihood.data, lower)), max(np.append(self.likelihood.data, upper))
             ymin, ymax = ymin - 0.1 * (ymax - ymin), ymax + 0.1 * (ymax - ymin)
             pb.xlim(xmin, xmax)
@@ -268,8 +273,7 @@ class GP(model):
             if hasattr(self, 'Z'):
                 Zu = self.Z * self._Xstd + self._Xmean
                 pb.plot(Zu, Zu * 0 + pb.ylim()[0], 'r|', mew=1.5, markersize=12)
-                if self.has_uncertain_inputs:
+                    # pb.errorbar(self.X[:,0], pb.ylim()[0]+np.zeros(self.N), xerr=2*np.sqrt(self.X_variance.flatten()))
-                    pb.errorbar(self.X[:,0], pb.ylim()[0]+np.zeros(self.N), xerr=2*np.sqrt(self.X_variance.flatten()))
 
         elif self.X.shape[1] == 2:  # FIXME
             resolution = resolution or 50