mirror of
https://github.com/SheffieldML/GPy.git
synced 2026-06-05 14:55:15 +02:00
REVERT a53690ab7f, flapack back substitued in
This commit is contained in:
Max Zwiessele
parent
43c2f8af87
commit
4eebf99bfb
15 changed files with 96 additions and 101 deletions
|
|
@ -95,7 +95,7 @@ class Bayesian_GPLVM(sparse_GP, GPLVM):
|
|||
return x
|
||||
|
||||
def _clipped(self, x):
|
||||
return np.clip(x, -1e300, 1e300)
|
||||
return x # np.clip(x, -1e300, 1e300)
|
||||
|
||||
def _set_params(self, x, save_old=True, save_count=0):
|
||||
# try:
|
||||
|
|
|
|||
|
|
@ -11,8 +11,8 @@ from sparse_GP import sparse_GP
|
|||
|
||||
def backsub_both_sides(L,X):
|
||||
""" Return L^-T * X * L^-1, assumuing X is symmetrical and L is lower cholesky"""
|
||||
tmp, _ = linalg.lapack.dtrtrs(L, np.asfortranarray(X), lower=1, trans=1)
|
||||
return linalg.lapack.dtrtrs(L, np.asfortranarray(tmp.T), lower=1, trans=1)[0].T
|
||||
tmp,_ = linalg.lapack.flapack.dtrtrs(L,np.asfortranarray(X),lower=1,trans=1)
|
||||
return linalg.lapack.flapack.dtrtrs(L,np.asfortranarray(tmp.T),lower=1,trans=1)[0].T
|
||||
|
||||
class FITC(sparse_GP):
|
||||
|
||||
|
|
@ -36,7 +36,7 @@ class FITC(sparse_GP):
|
|||
|
||||
#factor Kmm
|
||||
self.Lm = jitchol(self.Kmm)
|
||||
self.Lmi, info = linalg.lapack.dtrtrs(self.Lm, np.eye(self.M), lower=1)
|
||||
self.Lmi,info = linalg.lapack.flapack.dtrtrs(self.Lm,np.eye(self.M),lower=1)
|
||||
Lmipsi1 = np.dot(self.Lmi,self.psi1)
|
||||
self.Qnn = np.dot(Lmipsi1.T,Lmipsi1).copy()
|
||||
self.Diag0 = self.psi0 - np.diag(self.Qnn)
|
||||
|
|
@ -50,7 +50,7 @@ class FITC(sparse_GP):
|
|||
if self.likelihood.is_heteroscedastic:
|
||||
assert self.likelihood.D == 1
|
||||
tmp = self.psi1 * (np.sqrt(self.beta_star.flatten().reshape(1, self.N)))
|
||||
tmp, _ = linalg.lapack.dtrtrs(self.Lm, np.asfortranarray(tmp), lower=1)
|
||||
tmp, _ = linalg.lapack.flapack.dtrtrs(self.Lm, np.asfortranarray(tmp), lower=1)
|
||||
self.A = tdot(tmp)
|
||||
|
||||
# factor B
|
||||
|
|
@ -59,8 +59,8 @@ class FITC(sparse_GP):
|
|||
self.LBi = chol_inv(self.LB)
|
||||
self.psi1V = np.dot(self.psi1, self.V_star)
|
||||
|
||||
Lmi_psi1V, info = linalg.lapack.dtrtrs(self.Lm, np.asfortranarray(self.psi1V), lower=1, trans=0)
|
||||
self._LBi_Lmi_psi1V, _ = linalg.lapack.dtrtrs(self.LB, np.asfortranarray(Lmi_psi1V), lower=1, trans=0)
|
||||
Lmi_psi1V, info = linalg.lapack.flapack.dtrtrs(self.Lm, np.asfortranarray(self.psi1V), lower=1, trans=0)
|
||||
self._LBi_Lmi_psi1V, _ = linalg.lapack.flapack.dtrtrs(self.LB, np.asfortranarray(Lmi_psi1V), lower=1, trans=0)
|
||||
|
||||
Kmmipsi1 = np.dot(self.Lmi.T,Lmipsi1)
|
||||
b_psi1_Ki = self.beta_star * Kmmipsi1.T
|
||||
|
|
@ -190,7 +190,7 @@ class FITC(sparse_GP):
|
|||
self.P = Iplus_Dprod_i[:,None] * self.psi1.T
|
||||
self.RPT0 = np.dot(self.Lmi,self.psi1)
|
||||
self.L = np.linalg.cholesky(np.eye(self.M) + np.dot(self.RPT0,((1. - Iplus_Dprod_i)/self.Diag0)[:,None]*self.RPT0.T))
|
||||
self.R, info = linalg.dtrtrs(self.L, self.Lmi, lower=1)
|
||||
self.R,info = linalg.flapack.dtrtrs(self.L,self.Lmi,lower=1)
|
||||
self.RPT = np.dot(self.R,self.P.T)
|
||||
self.Sigma = np.diag(self.Diag) + np.dot(self.RPT.T,self.RPT)
|
||||
self.w = self.Diag * self.likelihood.v_tilde
|
||||
|
|
@ -212,7 +212,7 @@ class FITC(sparse_GP):
|
|||
# = I - [RPT0] * (U*U.T)^-1 * [RPT0].T
|
||||
# = I - V.T * V
|
||||
U = np.linalg.cholesky(np.diag(self.Diag0) + self.Qnn)
|
||||
V, info = linalg.dtrtrs(U, self.RPT0.T, lower=1)
|
||||
V,info = linalg.flapack.dtrtrs(U,self.RPT0.T,lower=1)
|
||||
C = np.eye(self.M) - np.dot(V.T,V)
|
||||
mu_u = np.dot(C,self.RPT0)*(1./self.Diag0[None,:])
|
||||
#self.C = C
|
||||
|
|
|
|||
|
|
@ -74,13 +74,13 @@ class GP(model):
|
|||
# the gradient of the likelihood wrt the covariance matrix
|
||||
if self.likelihood.YYT is None:
|
||||
#alpha = np.dot(self.Ki, self.likelihood.Y)
|
||||
alpha, _ = linalg.lapack.dpotrs(self.L, self.likelihood.Y, lower=1)
|
||||
alpha,_ = linalg.lapack.flapack.dpotrs(self.L, self.likelihood.Y,lower=1)
|
||||
|
||||
self.dL_dK = 0.5 * (tdot(alpha) - self.D * self.Ki)
|
||||
else:
|
||||
#tmp = mdot(self.Ki, self.likelihood.YYT, self.Ki)
|
||||
tmp, _ = linalg.lapack.dpotrs(self.L, np.asfortranarray(self.likelihood.YYT), lower=1)
|
||||
tmp, _ = linalg.lapack.dpotrs(self.L, np.asfortranarray(tmp.T), lower=1)
|
||||
tmp, _ = linalg.lapack.flapack.dpotrs(self.L, np.asfortranarray(self.likelihood.YYT), lower=1)
|
||||
tmp, _ = linalg.lapack.flapack.dpotrs(self.L, np.asfortranarray(tmp.T), lower=1)
|
||||
self.dL_dK = 0.5 * (tmp - self.D * self.Ki)
|
||||
|
||||
def _get_params(self):
|
||||
|
|
@ -104,7 +104,7 @@ class GP(model):
|
|||
Computes the model fit using YYT if it's available
|
||||
"""
|
||||
if self.likelihood.YYT is None:
|
||||
tmp, _ = linalg.lapack.dtrtrs(self.L, np.asfortranarray(self.likelihood.Y), lower=1)
|
||||
tmp, _ = linalg.lapack.flapack.dtrtrs(self.L, np.asfortranarray(self.likelihood.Y), lower=1)
|
||||
return -0.5 * np.sum(np.square(tmp))
|
||||
#return -0.5 * np.sum(np.square(np.dot(self.Li, self.likelihood.Y)))
|
||||
else:
|
||||
|
|
@ -136,7 +136,7 @@ class GP(model):
|
|||
"""
|
||||
Kx = self.kern.K(_Xnew,self.X,which_parts=which_parts).T
|
||||
#KiKx = np.dot(self.Ki, Kx)
|
||||
KiKx, _ = linalg.lapack.dpotrs(self.L, np.asfortranarray(Kx), lower=1)
|
||||
KiKx, _ = linalg.lapack.flapack.dpotrs(self.L, np.asfortranarray(Kx), lower=1)
|
||||
mu = np.dot(KiKx.T, self.likelihood.Y)
|
||||
if full_cov:
|
||||
Kxx = self.kern.K(_Xnew, which_parts=which_parts)
|
||||
|
|
|
|||
|
|
@ -11,8 +11,8 @@ from sparse_GP import sparse_GP
|
|||
|
||||
def backsub_both_sides(L,X):
|
||||
""" Return L^-T * X * L^-1, assumuing X is symmetrical and L is lower cholesky"""
|
||||
tmp, _ = linalg.lapack.dtrtrs(L, np.asfortranarray(X), lower=1, trans=1)
|
||||
return linalg.lapack.dtrtrs(L, np.asfortranarray(tmp.T), lower=1, trans=1)[0].T
|
||||
tmp,_ = linalg.lapack.flapack.dtrtrs(L,np.asfortranarray(X),lower=1,trans=1)
|
||||
return linalg.lapack.flapack.dtrtrs(L,np.asfortranarray(tmp.T),lower=1,trans=1)[0].T
|
||||
|
||||
|
||||
class generalized_FITC(sparse_GP):
|
||||
|
|
@ -82,7 +82,7 @@ class generalized_FITC(sparse_GP):
|
|||
if self.likelihood.is_heteroscedastic:
|
||||
|
||||
# Compute generalized FITC's diagonal term of the covariance
|
||||
self.Lmi, info = linalg.lapack.dtrtrs(self.Lm, np.eye(self.M), lower=1)
|
||||
self.Lmi,info = linalg.lapack.flapack.dtrtrs(self.Lm,np.eye(self.M),lower=1)
|
||||
Lmipsi1 = np.dot(self.Lmi,self.psi1)
|
||||
self.Qnn = np.dot(Lmipsi1.T,Lmipsi1)
|
||||
#self.Kmmi, Lm, Lmi, Kmm_logdet = pdinv(self.Kmm)
|
||||
|
|
@ -95,7 +95,7 @@ class generalized_FITC(sparse_GP):
|
|||
self.P = Iplus_Dprod_i[:,None] * self.psi1.T
|
||||
self.RPT0 = np.dot(self.Lmi,self.psi1)
|
||||
self.L = np.linalg.cholesky(np.eye(self.M) + np.dot(self.RPT0,((1. - Iplus_Dprod_i)/self.Diag0)[:,None]*self.RPT0.T))
|
||||
self.R, info = linalg.dtrtrs(self.L, self.Lmi, lower=1)
|
||||
self.R,info = linalg.flapack.dtrtrs(self.L,self.Lmi,lower=1)
|
||||
self.RPT = np.dot(self.R,self.P.T)
|
||||
self.Sigma = np.diag(self.Diag) + np.dot(self.RPT.T,self.RPT)
|
||||
self.w = self.Diag * self.likelihood.v_tilde
|
||||
|
|
@ -182,7 +182,7 @@ class generalized_FITC(sparse_GP):
|
|||
# = I - [RPT0] * (U*U.T)^-1 * [RPT0].T
|
||||
# = I - V.T * V
|
||||
U = np.linalg.cholesky(np.diag(self.Diag0) + self.Qnn)
|
||||
V, info = linalg.dtrtrs(U, self.RPT0.T, lower=1)
|
||||
V,info = linalg.flapack.dtrtrs(U,self.RPT0.T,lower=1)
|
||||
C = np.eye(self.M) - np.dot(V.T,V)
|
||||
mu_u = np.dot(C,self.RPT0)*(1./self.Diag0[None,:])
|
||||
#self.C = C
|
||||
|
|
|
|||
|
|
@ -65,7 +65,7 @@ class MRD(model):
|
|||
self._init = True
|
||||
X = self._init_X(initx, likelihood_or_Y_list)
|
||||
Z = self._init_Z(initz, X)
|
||||
self.bgplvms = [Bayesian_GPLVM(l, Q=Q, kernel=k, X=X, Z=Z, M=self.M, _debug=_debug, ** kw) for l, k in zip(likelihood_or_Y_list, kernels)]
|
||||
self.bgplvms = [Bayesian_GPLVM(l, Q=Q, kernel=k, X=X, Z=Z, M=self.M, **kw) for l, k in zip(likelihood_or_Y_list, kernels)]
|
||||
del self._init
|
||||
|
||||
self.gref = self.bgplvms[0]
|
||||
|
|
@ -143,9 +143,9 @@ class MRD(model):
|
|||
# S_names = sum([['X_variance_%i_%i' % (n, q) for q in range(self.Q)] for n in range(self.N)], [])
|
||||
n1 = self.gref._get_param_names()
|
||||
n1var = n1[:self.NQ * 2 + self.MQ]
|
||||
map_names = lambda ns, cd48_name: map(lambda x: "{1}_{0}".format(*x),
|
||||
map_names = lambda ns, name: map(lambda x: "{1}_{0}".format(*x),
|
||||
itertools.izip(ns,
|
||||
itertools.repeat(cd48_name)))
|
||||
itertools.repeat(name)))
|
||||
return list(itertools.chain(n1var, *(map_names(\
|
||||
sparse_GP._get_param_names(g)[self.MQ:], n) \
|
||||
for g, n in zip(self.bgplvms, self.names))))
|
||||
|
|
@ -213,12 +213,12 @@ class MRD(model):
|
|||
dLdmuS = numpy.hstack((dLdmu.flatten(), dLdS.flatten())).flatten()
|
||||
dldzt1 = reduce(lambda a, b: a + b, (sparse_GP._log_likelihood_gradients(g)[:self.MQ] for g in self.bgplvms))
|
||||
|
||||
return self.gref._clipped(numpy.hstack((dLdmuS,
|
||||
return numpy.hstack((dLdmuS,
|
||||
dldzt1,
|
||||
numpy.hstack([numpy.hstack([g.dL_dtheta(),
|
||||
g.likelihood._gradients(\
|
||||
partial=g.partial_for_likelihood)]) \
|
||||
for g in self.bgplvms]))))
|
||||
for g in self.bgplvms])))
|
||||
|
||||
def _init_X(self, init='PCA', likelihood_list=None):
|
||||
if likelihood_list is None:
|
||||
|
|
|
|||
|
|
@ -80,7 +80,7 @@ class sparse_GP(GP):
|
|||
tmp = self.psi1 * (np.sqrt(self.likelihood.precision.flatten().reshape(1, self.N)))
|
||||
else:
|
||||
tmp = self.psi1 * (np.sqrt(self.likelihood.precision))
|
||||
tmp, _ = linalg.lapack.dtrtrs(self.Lm, np.asfortranarray(tmp), lower=1)
|
||||
tmp, _ = linalg.lapack.flapack.dtrtrs(self.Lm, np.asfortranarray(tmp), lower=1)
|
||||
self.A = tdot(tmp)
|
||||
|
||||
|
||||
|
|
@ -92,10 +92,10 @@ class sparse_GP(GP):
|
|||
self.psi1V = np.dot(self.psi1, self.likelihood.V)
|
||||
|
||||
# back substutue C into psi1V
|
||||
tmp, info1 = linalg.lapack.dtrtrs(self.Lm, np.asfortranarray(self.psi1V), lower=1, trans=0)
|
||||
self._LBi_Lmi_psi1V, _ = linalg.lapack.dtrtrs(self.LB, np.asfortranarray(tmp), lower=1, trans=0)
|
||||
tmp, info2 = linalg.lapack.dpotrs(self.LB, tmp, lower=1)
|
||||
self.Cpsi1V, info3 = linalg.lapack.dtrtrs(self.Lm, tmp, lower=1, trans=1)
|
||||
tmp, info1 = linalg.lapack.flapack.dtrtrs(self.Lm, np.asfortranarray(self.psi1V), lower=1, trans=0)
|
||||
self._LBi_Lmi_psi1V, _ = linalg.lapack.flapack.dtrtrs(self.LB, np.asfortranarray(tmp), lower=1, trans=0)
|
||||
tmp, info2 = linalg.lapack.flapack.dpotrs(self.LB, tmp, lower=1)
|
||||
self.Cpsi1V, info3 = linalg.lapack.flapack.dtrtrs(self.Lm, tmp, lower=1, trans=1)
|
||||
|
||||
# Compute dL_dKmm
|
||||
tmp = tdot(self._LBi_Lmi_psi1V)
|
||||
|
|
@ -220,7 +220,7 @@ class sparse_GP(GP):
|
|||
def _raw_predict(self, Xnew, X_variance_new=None, which_parts='all', full_cov=False):
|
||||
"""Internal helper function for making predictions, does not account for normalization"""
|
||||
|
||||
Bi, _ = linalg.lapack.dpotri(self.LB, lower=0) # WTH? this lower switch should be 1, but that doesn't work!
|
||||
Bi, _ = linalg.lapack.flapack.dpotri(self.LB, lower=0) # WTH? this lower switch should be 1, but that doesn't work!
|
||||
symmetrify(Bi)
|
||||
Kmmi_LmiBLmi = backsub_both_sides(self.Lm, np.eye(self.M) - Bi)
|
||||
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue