[missing_data in sparse gp] can be extended towards missing_data handling in gp itself. Setting up gpy issue

GPy/core/parameterization/param.py

@@ -53,7 +53,9 @@ class Param(Parameterizable, ObsAr):
         return obj
 
     def __init__(self, name, input_array, default_constraint=None, *a, **kw):
+        self._in_init_ = True
         super(Param, self).__init__(name=name, default_constraint=default_constraint, *a, **kw)
+        self._in_init_ = False
 
     def build_pydot(self,G):
         import pydot

GPy/core/parameterization/parameter_core.py

@@ -18,7 +18,7 @@ import numpy as np
 import re
 import logging
 
-__updated__ = '2014-09-22'
+__updated__ = '2014-10-09'
 
 class HierarchyError(Exception):
     """
@@ -99,7 +99,7 @@ class Observable(object):
 
         :param bool trigger_parent: Whether to trigger the parent, after self has updated
         """
-        if not self.update_model():
+        if not self.update_model() or self._in_init_:
             #print "Warning: updates are off, updating the model will do nothing"
             return
         self._trigger_params_changed(trigger_parent)

GPy/core/sparse_gp.py

@@ -218,6 +218,7 @@ class SparseGP(GP):
         print message,
         for d in xrange(self.output_dim):
             ninan = self.ninan[:, d]
+
             print ' '*(len(message)) + '\r',
             message = m_f(d)
             print message,
@@ -249,9 +250,8 @@ class SparseGP(GP):
         if self.missing_data:
             self._outer_loop_for_missing_data()
         else:
-            self.posterior, self._log_marginal_likelihood, self.grad_dict, gradients, _ = self._inner_parameters_changed(self.kern, self.X, self.Z, self.likelihood, self.Y_normalized, self.Y_metadata)
+            self.posterior, self._log_marginal_likelihood, self.grad_dict, full_values, _ = self._inner_parameters_changed(self.kern, self.X, self.Z, self.likelihood, self.Y_normalized, self.Y_metadata)
-            self.kern.gradient = gradients['kerngrad']
+            self._outer_values_update(full_values)
-            self.Z.gradient = gradients['Zgrad']
 
     def _raw_predict(self, Xnew, full_cov=False, kern=None):
         """

GPy/examples/dimensionality_reduction.py

@@ -374,9 +374,6 @@ def bgplvm_simulation_missing_data(optimize=True, verbose=1,
     m = BayesianGPLVM(Ymissing, Q, init="random", num_inducing=num_inducing,
                       kernel=k, missing_data=True)
 
-    m.X.variance[:] = _np.random.uniform(0,.1,m.X.shape)
-    m.likelihood.variance = .01
-    m.parameters_changed()
     m.Yreal = Y
 
     if optimize:

GPy/models/bayesian_gplvm.py

@@ -32,7 +32,7 @@ class BayesianGPLVM(SparseGP_MPI):
         self.__IN_OPTIMIZATION__ = False
 
         self.logger = logging.getLogger(self.__class__.__name__)
-        if X == None:
+        if X is None:
             from ..util.initialization import initialize_latent
             self.logger.info("initializing latent space X with method {}".format(init))
             X, fracs = initialize_latent(init, input_dim, Y)
@@ -97,12 +97,17 @@ class BayesianGPLVM(SparseGP_MPI):
                                             Z=Z, dL_dpsi0=grad_dict['dL_dpsi0'],
                                             dL_dpsi1=grad_dict['dL_dpsi1'],
                                             dL_dpsi2=grad_dict['dL_dpsi2'])
+
+        # Subsetting Variational Posterior objects, makes the gradients
+        # empty. We need them to be 0 though:
         X.mean.gradient[:] = 0
         X.variance.gradient[:] = 0
+
         self.variational_prior.update_gradients_KL(X)
         current_values['meangrad'] += X.mean.gradient
         current_values['vargrad'] += X.variance.gradient
 
+        if subset_indices is not None:
             value_indices['meangrad'] = subset_indices['samples']
             value_indices['vargrad'] = subset_indices['samples']
         return posterior, log_marginal_likelihood, grad_dict, current_values, value_indices