apunkt/GPy
1
0
Fork 0
mirror of https://github.com/SheffieldML/GPy.git synced 2026-05-15 06:52:39 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

some messing with the linear algebra in sparse_GP. This should be more efficient... let's hope nothing breaks

Browse source
This commit is contained in:
James Hensman 2013-03-11 19:43:46 +00:00
parent 9b8c4eae25
commit d3c87feffa
1 changed files with 8 additions and 9 deletions
Download patch file Download diff file Expand all files Collapse all files

17
GPy/models/sparse_GP.py
View file

@ -102,17 +102,16 @@ class sparse_GP(GP):
self.Bi, self.LB, self.LBi, self.B_logdet = pdinv(self.B)
self.psi1V = np.dot(self.psi1, self.V)
self.psi1VVpsi1 = np.dot(self.psi1V, self.psi1V.T)
tmp = np.dot(self.Lmi.T, self.LBi.T)
self.C = np.dot(tmp,tmp.T)
#self.C = mdot(self.Lmi.T, self.Bi, self.Lmi)
#self.E = mdot(self.C, self.psi1VVpsi1/sf2, self.C.T)
tmp = np.dot(self.C/sf,self.psi1V)
self.E = np.dot(tmp,tmp.T)
self.Cpsi1V = np.dot(self.C,self.psi1V)
self.Cpsi1VVpsi1 = np.dot(self.Cpsi1V,self.psi1V.T)
self.E = np.dot(self.Cpsi1VVpsi1,self.C)/sf2
# Compute dL_dpsi # FIXME: this is untested for the heterscedastic + uncertin inputs case
self.dL_dpsi0 = - 0.5 * self.D * (self.likelihood.precision * np.ones([self.N,1])).flatten()
self.dL_dpsi1 = mdot(self.V, self.psi1V.T,self.C).T
#self.dL_dpsi1 = mdot(self.V, self.psi1V.T,self.C).T
self.dL_dpsi1 = np.dot(self.Cpsi1V,self.V.T)
if self.likelihood.is_heteroscedastic:
if self.has_uncertain_inputs:
self.dL_dpsi2 = 0.5 * self.likelihood.precision[:,None,None] * self.D * self.Kmmi[None,:,:] # dB
@ -139,7 +138,7 @@ class sparse_GP(GP):
# Compute dL_dKmm
self.dL_dKmm = -0.5 * self.D * mdot(self.Lmi.T, self.A, self.Lmi)*sf2 # dB
self.dL_dKmm += -0.5 * self.D * (- self.C/sf2 - 2.*mdot(self.C, self.psi2_beta_scaled, self.Kmmi) + self.Kmmi) # dC
self.dL_dKmm += np.dot(np.dot(self.E*sf2, self.psi2_beta_scaled) - np.dot(self.C, self.psi1VVpsi1), self.Kmmi) + 0.5*self.E # dD
self.dL_dKmm += np.dot(np.dot(self.E*sf2, self.psi2_beta_scaled) - self.Cpsi1VVpsi1, self.Kmmi) + 0.5*self.E # dD
#the partial derivative vector for the likelihood
if self.likelihood.Nparams ==0:
@ -157,7 +156,7 @@ class sparse_GP(GP):
dbeta = 0.5 * self.N*self.D/beta - 0.5 * np.sum(np.square(self.likelihood.Y))
dbeta += - 0.5 * self.D * (self.psi0.sum() - np.trace(self.A)/beta*sf2)
dbeta += - 0.5 * self.D * trace_dot(self.Bi,self.A)/beta
dbeta += np.sum((self.C - 0.5 * mdot(self.C,self.psi2_beta_scaled,self.C) ) * self.psi1VVpsi1 )/beta
dbeta += np.trace(self.Cpsi1VVpsi1)/beta - 0.5 * trace_dot(np.dot(self.C,self.psi2_beta_scaled) , self.Cpsi1VVpsi1 )/beta
self.partial_for_likelihood = -dbeta*self.likelihood.precision**2
@ -198,7 +197,7 @@ class sparse_GP(GP):
A = -0.5*self.N*self.D*(np.log(2.*np.pi) - np.log(self.likelihood.precision)) -0.5*self.likelihood.precision*self.likelihood.trYYT
B = -0.5*self.D*(np.sum(self.likelihood.precision*self.psi0) - np.trace(self.A)*sf2)
C = -0.5*self.D * (self.B_logdet + self.M*np.log(sf2))
D = +0.5*np.sum(self.psi1VVpsi1 * self.C)
D = 0.5*np.trace(self.Cpsi1VVpsi1)
return A+B+C+D
def _log_likelihood_gradients(self):