apunkt/GPy
1
0
Fork 0
mirror of https://github.com/SheffieldML/GPy.git synced 2026-05-21 14:05:14 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

tidied the computation of A in sparse_GP

Browse source
This commit is contained in:
James Hensman 2013-05-19 19:17:55 +01:00
parent 58b206a245
commit 3f22e61d2d
1 changed files with 12 additions and 29 deletions
Download patch file Download diff file Expand all files Collapse all files

31
GPy/models/sparse_GP.py
View file

@ -70,40 +70,23 @@ class sparse_GP(GP):
self.Lm = jitchol(self.Kmm) self.Lm = jitchol(self.Kmm)
# The rather complex computations of self.A # The rather complex computations of self.A
if self.has_uncertain_inputs:
if self.likelihood.is_heteroscedastic: if self.likelihood.is_heteroscedastic:
assert self.likelihood.D == 1 # TODO: what if the likelihood is heterscedatic and there are multiple independent outputs? psi2_beta = (self.psi2 * (self.likelihood.precision.flatten().reshape(self.N, 1, 1))).sum(0)
if self.has_uncertain_inputs: else:
# psi2_beta_scaled = (self.psi2 * (self.likelihood.precision.flatten().reshape(self.N, 1, 1) / sf2)).sum(0) psi2_beta = self.psi2.sum(0) * self.likelihood.precision
psi2_beta_scaled = (self.psi2 * (self.likelihood.precision.flatten().reshape(self.N, 1, 1))).sum(0) evals, evecs = linalg.eigh(psi2_beta)
evals, evecs = linalg.eigh(psi2_beta_scaled)
clipped_evals = np.clip(evals, 0., 1e6) # TODO: make clipping configurable clipped_evals = np.clip(evals, 0., 1e6) # TODO: make clipping configurable
if not np.allclose(evals, clipped_evals):
print "Warning: clipping posterior eigenvalues"
tmp = evecs * np.sqrt(clipped_evals) tmp = evecs * np.sqrt(clipped_evals)
tmp, _ = linalg.lapack.flapack.dtrtrs(self.Lm, np.asfortranarray(tmp), lower=1)
self.A = tdot(tmp)
else: else:
# tmp = self.psi1 * (np.sqrt(self.likelihood.precision.flatten().reshape(1, self.N)) / sf) if self.likelihood.is_heteroscedastic:
tmp = self.psi1 * (np.sqrt(self.likelihood.precision.flatten().reshape(1, self.N))) tmp = self.psi1 * (np.sqrt(self.likelihood.precision.flatten().reshape(1, self.N)))
tmp, _ = linalg.lapack.flapack.dtrtrs(self.Lm, np.asfortranarray(tmp), lower=1)
self.A = tdot(tmp)
else: else:
if self.has_uncertain_inputs:
# psi2_beta_scaled = (self.psi2 * (self.likelihood.precision / sf2)).sum(0)
psi2_beta_scaled = (self.psi2 * (self.likelihood.precision)).sum(0)
evals, evecs = linalg.eigh(psi2_beta_scaled)
clipped_evals = np.clip(evals, 0., 1e15) # TODO: make clipping configurable
if not np.allclose(evals, clipped_evals):
print "Warning: clipping posterior eigenvalues"
tmp = evecs * np.sqrt(clipped_evals)
tmp, _ = linalg.lapack.flapack.dtrtrs(self.Lm, np.asfortranarray(tmp), lower=1)
self.A = tdot(tmp)
else:
# tmp = self.psi1 * (np.sqrt(self.likelihood.precision) / sf)
tmp = self.psi1 * (np.sqrt(self.likelihood.precision)) tmp = self.psi1 * (np.sqrt(self.likelihood.precision))
tmp, _ = linalg.lapack.flapack.dtrtrs(self.Lm, np.asfortranarray(tmp), lower=1) tmp, _ = linalg.lapack.flapack.dtrtrs(self.Lm, np.asfortranarray(tmp), lower=1)
self.A = tdot(tmp) self.A = tdot(tmp)
# factor B # factor B
# self.B = np.eye(self.M) / sf2 + self.A # self.B = np.eye(self.M) / sf2 + self.A
self.B = np.eye(self.M) + self.A self.B = np.eye(self.M) + self.A