apunkt/GPy
1
0
Fork 0
mirror of https://github.com/SheffieldML/GPy.git synced 2026-05-30 14:35:15 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

Seem to have gradients much closer now

Browse source
This commit is contained in:
Alan Saul 2013-05-08 16:05:01 +01:00
parent 84f12c1079
commit 6c4866662c
4 changed files with 110 additions and 60 deletions
Download patch file Download diff file Expand all files Collapse all files

18
GPy/models/GP.py
View file

@ -125,19 +125,23 @@ class GP(model):
if isinstance(self.likelihood, Laplace):
dL_dthetaK_explicit = dL_dthetaK
#Need to pass in a matrix of ones to get access to raw dK_dthetaK values without being chained
fake_dL_dKs = np.ones(self.dL_dK.shape)
fake_dL_dKs = np.eye(self.dL_dK.shape[0])
dK_dthetaK = self.kern.dK_dtheta(dL_dK=fake_dL_dKs, X=self.X)
dL_dthetaK_implicit = self.likelihood._Kgradients(self.dL_dK, dK_dthetaK)
#We need the dL_dK where K is equal to the prior K, not K+Sigma as is the case now
dL_dthetaK_implicit = self.likelihood._Kgradients(dL_d_K_Sigma=self.dL_dK, dK_dthetaK=dK_dthetaK)
dL_dthetaK = dL_dthetaK_explicit + dL_dthetaK_implicit
print "dL_dthetaK_explicit: {dldkx} dL_dthetaK_implicit: {dldki} dL_dthetaK: {dldk}".format(dldkx=dL_dthetaK_explicit, dldki=dL_dthetaK_implicit, dldk=dL_dthetaK)
#print "dL_dthetaK_explicit: {dldkx} dL_dthetaK_implicit: {dldki} dL_dthetaK: {dldk}".format(dldkx=dL_dthetaK_explicit, dldki=dL_dthetaK_implicit, dldk=dL_dthetaK)
dL_dthetaL = self.likelihood._gradients(partial=self.dL_dK)
else:
print "dL_dthetaK: ", dL_dthetaK
dL_dthetaL = self.likelihood._gradients(partial=np.diag(self.dL_dK))
print "dL_dthetaL: ", dL_dthetaL
print "Stacked dL_dthetaK, dL_dthetaL: ", np.hstack((dL_dthetaK, dL_dthetaL))
import ipdb; ipdb.set_trace() ### XXX BREAKPOINT
else:
#print "dL_dthetaK: ", dL_dthetaK
dL_dthetaL = self.likelihood._gradients(partial=np.diag(self.dL_dK))
print "Stacked dL_dthetaK, dL_dthetaL: ", np.hstack((dL_dthetaK, dL_dthetaL))
#print "dL_dthetaL: ", dL_dthetaL
return np.hstack((dL_dthetaK, dL_dthetaL))
#return np.hstack((self.kern.dK_dtheta(dL_dK=self.dL_dK, X=self.X), self.likelihood._gradients(partial=np.diag(self.dL_dK))))