apunkt/GPy
1
0
Fork 0
mirror of https://github.com/SheffieldML/GPy.git synced 2026-05-09 20:12:38 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

[minibatch] added coverage for branching, spottet bug in X_variance

Browse source
This commit is contained in:
mzwiessele 2016-03-08 11:19:50 +00:00
parent 88a9b92c80
commit 67043e83d7
2 changed files with 13 additions and 12 deletions
Download patch file Download diff file Expand all files Collapse all files

23
GPy/models/bayesian_gplvm_minibatch.py
View file

@ -40,12 +40,13 @@ class BayesianGPLVMMiniBatch(SparseGPMiniBatch):
Z = np.random.permutation(X.copy())[:num_inducing] Z = np.random.permutation(X.copy())[:num_inducing]
assert Z.shape[1] == X.shape[1] assert Z.shape[1] == X.shape[1]
if X_variance == False: if X_variance is False:
self.logger.info('no variance on X, activating sparse GPLVM') self.logger.info('no variance on X, activating sparse GPLVM')
X = Param("latent space", X) X = Param("latent space", X)
elif X_variance is None: else:
self.logger.info("initializing latent space variance ~ uniform(0,.1)") if X_variance is None:
X_variance = np.random.uniform(0,.1,X.shape) self.logger.info("initializing latent space variance ~ uniform(0,.1)")
X_variance = np.random.uniform(0,.1,X.shape)
self.variational_prior = NormalPrior() self.variational_prior = NormalPrior()
X = NormalPosterior(X, X_variance) X = NormalPosterior(X, X_variance)
@ -71,13 +72,13 @@ class BayesianGPLVMMiniBatch(SparseGPMiniBatch):
self.X = X self.X = X
self.link_parameter(self.X, 0) self.link_parameter(self.X, 0)
def set_X_gradients(self, X, X_grad): #def set_X_gradients(self, X, X_grad):
"""Set the gradients of the posterior distribution of X in its specific form.""" # """Set the gradients of the posterior distribution of X in its specific form."""
X.mean.gradient, X.variance.gradient = X_grad # X.mean.gradient, X.variance.gradient = X_grad
def get_X_gradients(self, X): #def get_X_gradients(self, X):
"""Get the gradients of the posterior distribution of X in its specific form.""" # """Get the gradients of the posterior distribution of X in its specific form."""
return X.mean.gradient, X.variance.gradient # return X.mean.gradient, X.variance.gradient
def _outer_values_update(self, full_values): def _outer_values_update(self, full_values):
""" """
@ -122,7 +123,7 @@ class BayesianGPLVMMiniBatch(SparseGPMiniBatch):
if self.missing_data or not self.stochastics: if self.missing_data or not self.stochastics:
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)
elif self.stochastics: else: #self.stochastics is given:
d = self.output_dim d = self.output_dim
self._log_marginal_likelihood -= kl_fctr*self.variational_prior.KL_divergence(self.X)*self.stochastics.batchsize/d self._log_marginal_likelihood -= kl_fctr*self.variational_prior.KL_divergence(self.X)*self.stochastics.batchsize/d

2
GPy/testing/minibatch_tests.py
View file

@ -54,7 +54,7 @@ class BGPLVMTest(unittest.TestCase):
def test_lik_comparisons_m0_s0(self): def test_lik_comparisons_m0_s0(self):
# Test if the different implementations give the exact same likelihood as the full model. # Test if the different implementations give the exact same likelihood as the full model.
# All of the following settings should give the same likelihood and gradients as the full model: # All of the following settings should give the same likelihood and gradients as the full model:
m = GPy.models.bayesian_gplvm_minibatch.BayesianGPLVMMiniBatch(self.Y, self.Q, missing_data=False, stochastic=False) m = GPy.models.bayesian_gplvm_minibatch.BayesianGPLVMMiniBatch(self.Y, self.Q, X_variance=self.m_full.X.variance.values, missing_data=False, stochastic=False)
m[:] = self.m_full[:] m[:] = self.m_full[:]
np.testing.assert_almost_equal(m.log_likelihood(), self.m_full.log_likelihood(), 7) np.testing.assert_almost_equal(m.log_likelihood(), self.m_full.log_likelihood(), 7)
np.testing.assert_allclose(m.gradient, self.m_full.gradient) np.testing.assert_allclose(m.gradient, self.m_full.gradient)