apunkt/GPy
1
0
Fork 0
mirror of https://github.com/SheffieldML/GPy.git synced 2026-05-10 20:42:39 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

Gradients for X seem to match

Browse source
This commit is contained in:
Alan Saul 2015-08-28 11:47:10 +03:00
parent afba8d5c5e
commit e12575c0c6
2 changed files with 41 additions and 10 deletions
Download patch file Download diff file Expand all files Collapse all files

49
GPy/models/bayesian_gplvm_minibatch.py
View file

@ -83,6 +83,35 @@ class BayesianGPLVMMiniBatch(SparseGPMiniBatch):
def _inner_parameters_changed(self, kern, X, Z, likelihood, Y, Y_metadata, Lm=None, dL_dKmm=None,psi0=None, psi1=None, psi2=None, **kw): def _inner_parameters_changed(self, kern, X, Z, likelihood, Y, Y_metadata, Lm=None, dL_dKmm=None,psi0=None, psi1=None, psi2=None, **kw):
posterior, log_marginal_likelihood, grad_dict = super(BayesianGPLVMMiniBatch, self)._inner_parameters_changed(kern, X, Z, likelihood, Y, Y_metadata, Lm=Lm, dL_dKmm=dL_dKmm, posterior, log_marginal_likelihood, grad_dict = super(BayesianGPLVMMiniBatch, self)._inner_parameters_changed(kern, X, Z, likelihood, Y, Y_metadata, Lm=Lm, dL_dKmm=dL_dKmm,
psi0=psi0, psi1=psi1, psi2=psi2, **kw) psi0=psi0, psi1=psi1, psi2=psi2, **kw)
kl_fctr = self.kl_factr
if self.has_uncertain_inputs():
if self.missing_data:
d = self.output_dim
log_marginal_likelihood -= kl_fctr*self.variational_prior.KL_divergence(X)/d
else:
log_marginal_likelihood -= kl_fctr*self.variational_prior.KL_divergence(X)
X.mean.gradient[:] = 0
X.variance.gradient[:] = 0
self.variational_prior.update_gradients_KL(X)
if self.missing_data:
grad_dict['meangrad'] = kl_fctr*X.mean.gradient/d
grad_dict['vargrad'] = kl_fctr*X.variance.gradient/d
else:
grad_dict['meangrad'] = kl_fctr*X.mean.gradient
grad_dict['vargrad'] = kl_fctr*X.variance.gradient
#if subset_indices is not None:
#value_indices['meangrad'] = subset_indices['samples']
#value_indices['vargrad'] = subset_indices['samples']
#grad_dict['meangrad'] = kl_fctr*self.X.mean.gradient
#grad_dict['vargrad'] = kl_fctr*self.X.variance.gradient
#self.X.mean.gradient[:] = 0
#self.X.variance.gradient[:] = 0
#import ipdb; ipdb.set_trace() # XXX BREAKPOINT
return posterior, log_marginal_likelihood, grad_dict return posterior, log_marginal_likelihood, grad_dict
def _outer_values_update(self, full_values): def _outer_values_update(self, full_values):
@ -92,28 +121,28 @@ class BayesianGPLVMMiniBatch(SparseGPMiniBatch):
""" """
super(BayesianGPLVMMiniBatch, self)._outer_values_update(full_values) super(BayesianGPLVMMiniBatch, self)._outer_values_update(full_values)
if self.has_uncertain_inputs(): if self.has_uncertain_inputs():
full_values['meangrad'], full_values['vargrad'] = self.kern.gradients_qX_expectations( meangrad_tmp, vargrad_tmp = self.kern.gradients_qX_expectations(
variational_posterior=self.X, variational_posterior=self.X,
Z=self.Z, dL_dpsi0=full_values['dL_dpsi0'], Z=self.Z, dL_dpsi0=full_values['dL_dpsi0'],
dL_dpsi1=full_values['dL_dpsi1'], dL_dpsi1=full_values['dL_dpsi1'],
dL_dpsi2=full_values['dL_dpsi2'], dL_dpsi2=full_values['dL_dpsi2'],
psi0=self.psi0, psi1=self.psi1, psi2=self.psi2) psi0=self.psi0, psi1=self.psi1, psi2=self.psi2)
full_values['meangrad'] += meangrad_tmp
full_values['vargrad'] += vargrad_tmp
else: else:
full_values['Xgrad'] = self.kern.gradients_X(full_values['dL_dKnm'], self.X, self.Z) full_values['Xgrad'] = self.kern.gradients_X(full_values['dL_dKnm'], self.X, self.Z)
full_values['Xgrad'] += self.kern.gradients_X_diag(full_values['dL_dKdiag'], self.X) full_values['Xgrad'] += self.kern.gradients_X_diag(full_values['dL_dKdiag'], self.X)
kl_fctr = self.kl_factr #kl_fctr = self.kl_factr
if self.has_uncertain_inputs(): #if self.has_uncertain_inputs():
self._log_marginal_likelihood -= kl_fctr*self.variational_prior.KL_divergence(self.X) #self._log_marginal_likelihood -= kl_fctr*self.variational_prior.KL_divergence(self.X)
# Subsetting Variational Posterior objects, makes the gradients # Subsetting Variational Posterior objects, makes the gradients
# empty. We need them to be 0 though: # empty. We need them to be 0 though:
self.X.mean.gradient[:] = 0 #self.X.mean.gradient[:] = 0
self.X.variance.gradient[:] = 0 #self.X.variance.gradient[:] = 0
self.variational_prior.update_gradients_KL(self.X) #self.variational_prior.update_gradients_KL(self.X)
full_values['meangrad'] += kl_fctr*self.X.mean.gradient
full_values['vargrad'] += kl_fctr*self.X.variance.gradient
if self.has_uncertain_inputs(): if self.has_uncertain_inputs():
self.X.mean.gradient = full_values['meangrad'] self.X.mean.gradient = full_values['meangrad']
@ -123,6 +152,8 @@ class BayesianGPLVMMiniBatch(SparseGPMiniBatch):
def _outer_init_full_values(self): def _outer_init_full_values(self):
full_values = super(BayesianGPLVMMiniBatch, self)._outer_init_full_values() full_values = super(BayesianGPLVMMiniBatch, self)._outer_init_full_values()
full_values['meangrad'] = np.zeros((self.X.shape[0], self.X.shape[1]))
full_values['vargrad'] = np.zeros((self.X.shape[0], self.X.shape[1]))
full_values['dL_dpsi0'] = np.zeros(self.X.shape[0]) full_values['dL_dpsi0'] = np.zeros(self.X.shape[0])
full_values['dL_dpsi1'] = np.zeros((self.X.shape[0], self.Z.shape[0])) full_values['dL_dpsi1'] = np.zeros((self.X.shape[0], self.Z.shape[0]))
return full_values return full_values

2
GPy/models/sparse_gp_minibatch.py
View file

@ -261,7 +261,7 @@ class SparseGPMiniBatch(SparseGP):
psi1ni = psi1[ninan] psi1ni = psi1[ninan]
if self.has_uncertain_inputs(): if self.has_uncertain_inputs():
psi2ni = psi2[ninan] psi2ni = psi2[ninan]
value_indices = dict(outputs=d, samples=ninan, dL_dpsi0=ninan, dL_dpsi1=ninan) value_indices = dict(outputs=d, samples=ninan, dL_dpsi0=ninan, dL_dpsi1=ninan, meangrad=ninan, vargrad=ninan)
else: else:
psi2ni = None psi2ni = None
value_indices = dict(outputs=d, samples=ninan, dL_dKdiag=ninan, dL_dKnm=ninan) value_indices = dict(outputs=d, samples=ninan, dL_dKdiag=ninan, dL_dKnm=ninan)