apunkt/GPy
1
0
Fork 0
mirror of https://github.com/SheffieldML/GPy.git synced 2026-05-18 13:55:14 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

Fixed log predictive density, added option for LOO to provide some intemediate variables

Browse source
This commit is contained in:
Alan Saul 2015-04-16 09:25:18 +01:00
parent fe0a4285ca
commit 361f0a5274
2 changed files with 22 additions and 13 deletions
Download patch file Download diff file Expand all files Collapse all files

18
GPy/inference/latent_function_inference/laplace.py
View file

@ -40,7 +40,7 @@ class Laplace(LatentFunctionInference):
self.first_run = True
self._previous_Ki_fhat = None
def LOO(self, kern, X, Y, likelihood, posterior, Y_metadata=None, K=None):
def LOO(self, kern, X, Y, likelihood, posterior, Y_metadata=None, K=None, f_hat=None, W=None, Ki_W_i=None):
"""
Leave one out log predictive density as found in
"Bayesian leave-one-out cross-validation approximations for Gaussian latent variable models"
@ -51,13 +51,19 @@ class Laplace(LatentFunctionInference):
if K is None:
K = kern.K(X)
f_hat, _ = self.rasm_mode(K, Y, likelihood, Ki_f_init, Y_metadata=Y_metadata)
W = -likelihood.d2logpdf_df2(f_hat, Y, Y_metadata=Y_metadata)
if f_hat is None:
f_hat, _ = self.rasm_mode(K, Y, likelihood, Ki_f_init, Y_metadata=Y_metadata)
if W is None:
W = -likelihood.d2logpdf_df2(f_hat, Y, Y_metadata=Y_metadata)
if Ki_W_i is None:
_, _, _, Ki_W_i = self._compute_B_statistics(K, W, likelihood.log_concave)
logpdf_dfhat = likelihood.dlogpdf_df(f_hat, Y, Y_metadata=Y_metadata)
K_Wi_i, _, _, Ki_W_i = self._compute_B_statistics(K, W, likelihood.log_concave)
W = np.diagflat(W)
if W.shape[1] == 1:
W = np.diagflat(W)
#Eq 14, and 16
var_site = 1./np.diag(W)[:, None]