mirror of
https://github.com/SheffieldML/GPy.git
synced 2026-06-08 15:05:15 +02:00
new shape for psi2
This commit is contained in:
Nicolo Fusi
parent
ff1b64022e
commit
10c774e84e
3 changed files with 19 additions and 19 deletions
|
|
@ -259,29 +259,29 @@ class kern(parameterised):
|
|||
:Z: np.ndarray of inducing inputs (M x Q)
|
||||
: mu, S: np.ndarrays of means and variacnes (each N x Q)
|
||||
:returns psi2: np.ndarray (N,M,M,Q) """
|
||||
target = np.zeros((Z.shape[0],Z.shape[0]))
|
||||
target = np.zeros((mu.shape[0],Z.shape[0],Z.shape[0]))
|
||||
slices1, slices2 = self._process_slices(slices1,slices2)
|
||||
[p.psi2(Z[s2,i_s],mu[s1,i_s],S[s1,i_s],target[s2,s2]) for p,i_s,s1,s2 in zip(self.parts,self.input_slices,slices1,slices2)]
|
||||
[p.psi2(Z[s2,i_s],mu[s1,i_s],S[s1,i_s],target[s1,s2,s2]) for p,i_s,s1,s2 in zip(self.parts,self.input_slices,slices1,slices2)]
|
||||
return target
|
||||
|
||||
def dpsi2_dtheta(self,partial,Z,mu,S,slices1=None,slices2=None):
|
||||
"""Returns shape (N,M,M,Ntheta)"""
|
||||
slices1, slices2 = self._process_slices(slices1,slices2)
|
||||
target = np.zeros(self.Nparam)
|
||||
[p.dpsi2_dtheta(partial[s2,s2],Z[s2,i_s],mu[s1,i_s],S[s1,i_s],target[ps]) for p,i_s,s1,s2,ps in zip(self.parts,self.input_slices,slices1,slices2,self.param_slices)]
|
||||
[p.dpsi2_dtheta(partial[s1,s2,s2],Z[s2,i_s],mu[s1,i_s],S[s1,i_s],target[ps]) for p,i_s,s1,s2,ps in zip(self.parts,self.input_slices,slices1,slices2,self.param_slices)]
|
||||
return target
|
||||
|
||||
def dpsi2_dZ(self,partial,Z,mu,S,slices1=None,slices2=None):
|
||||
slices1, slices2 = self._process_slices(slices1,slices2)
|
||||
target = np.zeros_like(Z)
|
||||
[p.dpsi2_dZ(partial[s2,s2],Z[s2,i_s],mu[s1,i_s],S[s1,i_s],target[s2,i_s]) for p,i_s,s1,s2 in zip(self.parts,self.input_slices,slices1,slices2)]
|
||||
[p.dpsi2_dZ(partial[s1,s2,s2],Z[s2,i_s],mu[s1,i_s],S[s1,i_s],target[s2,i_s]) for p,i_s,s1,s2 in zip(self.parts,self.input_slices,slices1,slices2)]
|
||||
return target
|
||||
|
||||
def dpsi2_dmuS(self,partial,Z,mu,S,slices1=None,slices2=None):
|
||||
"""return shapes are N,M,M,Q"""
|
||||
slices1, slices2 = self._process_slices(slices1,slices2)
|
||||
target_mu, target_S = np.zeros((2,mu.shape[0],mu.shape[1]))
|
||||
[p.dpsi2_dmuS(partial[s2,s2],Z[s2,i_s],mu[s1,i_s],S[s1,i_s],target_mu[s1,i_s],target_S[s1,i_s]) for p,i_s,s1,s2 in zip(self.parts,self.input_slices,slices1,slices2)]
|
||||
[p.dpsi2_dmuS(partial[s1,s2,s2],Z[s2,i_s],mu[s1,i_s],S[s1,i_s],target_mu[s1,i_s],target_S[s1,i_s]) for p,i_s,s1,s2 in zip(self.parts,self.input_slices,slices1,slices2)]
|
||||
|
||||
#TODO: there are some extra terms to compute here!
|
||||
return target_mu, target_S
|
||||
|
|
|
|||
|
|
@ -106,31 +106,31 @@ class rbf_ARD(kernpart):
|
|||
|
||||
def psi2(self,Z,mu,S,target):
|
||||
self._psi_computations(Z,mu,S)
|
||||
target += self._psi2.sum(0) #TODO: psi2 should be NxMxM (for het. noise)
|
||||
target += self._psi2
|
||||
|
||||
def dpsi2_dtheta(self,partial,Z,mu,S,target):
|
||||
"""Shape N,M,M,Ntheta"""
|
||||
self._psi_computations(Z,mu,S)
|
||||
d_var = np.sum(2.*self._psi2/self.variance,0)
|
||||
d_var = 2.*self._psi2/self.variance
|
||||
d_length = self._psi2[:,:,:,None]*(0.5*self._psi2_Zdist_sq*self._psi2_denom + 2.*self._psi2_mudist_sq + 2.*S[:,None,None,:]/self.lengthscales2)/(self.lengthscales*self._psi2_denom)
|
||||
d_length = d_length.sum(0)
|
||||
# d_length = d_length.sum(0)
|
||||
target[0] += np.sum(partial*d_var)
|
||||
target[1:] += (d_length*partial[:,:,None]).sum(0).sum(0)
|
||||
target[1:] += (d_length*partial[:,:,:,None]).sum(0).sum(0).sum(0)
|
||||
|
||||
def dpsi2_dZ(self,partial,Z,mu,S,target):
|
||||
"""Returns shape N,M,M,Q"""
|
||||
self._psi_computations(Z,mu,S)
|
||||
term1 = 0.5*self._psi2_Zdist/self.lengthscales2 # M, M, Q
|
||||
term2 = self._psi2_mudist/self._psi2_denom/self.lengthscales2 # N, M, M, Q
|
||||
dZ = self._psi2[:,:,:,None] * (term1[None] + term2)
|
||||
target += (partial[None,:,:,None]*dZ).sum(0).sum(0)
|
||||
dZ = self._psi2[:,:,:,None] * (term1[None] + term2)
|
||||
target += (partial[:,:,:,None]*dZ).sum(0).sum(0)
|
||||
|
||||
def dpsi2_dmuS(self,partial,Z,mu,S,target_mu,target_S):
|
||||
"""Think N,M,M,Q """
|
||||
self._psi_computations(Z,mu,S)
|
||||
tmp = self._psi2[:,:,:,None]/self.lengthscales2/self._psi2_denom
|
||||
target_mu += (partial[None,:,:,None]*-tmp*2.*self._psi2_mudist).sum(1).sum(1)
|
||||
target_S += (partial[None,:,:,None]*tmp*(2.*self._psi2_mudist_sq-1)).sum(1).sum(1)
|
||||
target_mu += (partial[:,:,:,None]*-tmp*2.*self._psi2_mudist).sum(1).sum(1)
|
||||
target_S += (partial[:,:,:,None]*tmp*(2.*self._psi2_mudist_sq-1)).sum(1).sum(1)
|
||||
|
||||
def _K_computations(self,X,X2):
|
||||
if not (np.all(X==self._X) and np.all(X2==self._X2)):
|
||||
|
|
|
|||
|
|
@ -70,7 +70,7 @@ class sparse_GP_regression(GP_regression):
|
|||
self.psi0 = self.kern.psi0(self.Z,self.X, self.X_uncertainty).sum()
|
||||
self.psi1 = self.kern.psi1(self.Z,self.X, self.X_uncertainty).T
|
||||
self.psi2 = self.kern.psi2(self.Z,self.X, self.X_uncertainty)
|
||||
self.psi2_beta_scaled = self.psi2*(self.beta/self.scale_factor**2)
|
||||
self.psi2_beta_scaled = (self.psi2*(self.beta/self.scale_factor**2)).sum(0)
|
||||
else:
|
||||
self.psi0 = self.kern.Kdiag(self.X,slices=self.Xslices).sum()
|
||||
self.psi1 = self.kern.K(self.Z,self.X)
|
||||
|
|
@ -98,9 +98,9 @@ class sparse_GP_regression(GP_regression):
|
|||
# Compute dL_dpsi
|
||||
self.dL_dpsi0 = - 0.5 * self.D * self.beta * np.ones(self.N)
|
||||
self.dL_dpsi1 = mdot(self.V, self.psi1V.T,self.C).T
|
||||
self.dL_dpsi2 = 0.5 * self.beta * self.D * self.Kmmi # dB
|
||||
self.dL_dpsi2 += - 0.5 * self.beta/sf2 * self.D * self.C # dC
|
||||
self.dL_dpsi2 += - 0.5 * self.beta * self.E # dD
|
||||
self.dL_dpsi2 = 0.5 * self.beta * self.D * self.Kmmi[None,:,:] # dB
|
||||
self.dL_dpsi2 += - 0.5 * self.beta/sf2 * self.D * self.C[None,:,:] # dC
|
||||
self.dL_dpsi2 += - 0.5 * self.beta * self.E[None,:,:] # dD
|
||||
|
||||
# Compute dL_dKmm
|
||||
self.dL_dKmm = -0.5 * self.D * mdot(self.Lmi.T, self.A, self.Lmi)*sf2 # dB
|
||||
|
|
@ -152,7 +152,7 @@ class sparse_GP_regression(GP_regression):
|
|||
dL_dtheta += self.kern.dpsi2_dtheta(self.dL_dpsi2,self.Z,self.X, self.X_uncertainty) # for multiple_beta, dL_dpsi2 will be a different shape
|
||||
else:
|
||||
#re-cast computations in psi2 back to psi1:
|
||||
dL_dpsi1 = self.dL_dpsi1 + 2.*np.dot(self.dL_dpsi2,self.psi1)
|
||||
dL_dpsi1 = self.dL_dpsi1 + 2.*np.dot(self.dL_dpsi2.sum(0),self.psi1)
|
||||
dL_dtheta += self.kern.dK_dtheta(dL_dpsi1,self.Z,self.X)
|
||||
dL_dtheta += self.kern.dKdiag_dtheta(self.dL_dpsi0, self.X)
|
||||
|
||||
|
|
@ -168,7 +168,7 @@ class sparse_GP_regression(GP_regression):
|
|||
dL_dZ += 2.*self.kern.dpsi2_dZ(self.dL_dpsi2,self.Z,self.X, self.X_uncertainty) # 'stripes'
|
||||
else:
|
||||
#re-cast computations in psi2 back to psi1:
|
||||
dL_dpsi1 = self.dL_dpsi1 + 2.*np.dot(self.dL_dpsi2,self.psi1)
|
||||
dL_dpsi1 = self.dL_dpsi1 + 2.*np.dot(self.dL_dpsi2.sum(0),self.psi1)
|
||||
dL_dZ += self.kern.dK_dX(dL_dpsi1,self.Z,self.X)
|
||||
return dL_dZ
|
||||
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue