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

UY dkdtheta

Browse source
This commit is contained in:
mu 2013-12-11 16:24:35 +00:00
parent 9b32fd47ee
commit c793e5d916
1 changed files with 47 additions and 13 deletions
Download patch file Download diff file Expand all files Collapse all files

60
GPy/kern/parts/ODE_UY.py
View file

@ -186,20 +186,29 @@ class ODE_UY(Kernpart):
def dK_dtheta(self, dL_dK, X, X2, target):
"""derivative of the covariance matrix with respect to the parameters."""
if X2 is None: X2 = X
dist = np.abs(X - X2.T)
X,slices = X[:,:-1],index_to_slices(X[:,-1])
if X2 is None:
X2,slices2 = X,slices
else:
X2,slices2 = X2[:,:-1],index_to_slices(X2[:,-1])
#rdist = X[:,0][:,None] - X2[:,0][:,None].T
rdist = X - X2.T
ly=1/self.lengthscaleY
lu=np.sqrt(3)/self.lengthscaleU
#ly=self.lengthscaleY
#lu=self.lengthscaleU
rd=rdist.shape[0]
dktheta1 = np.zeros([rd,rd])
dktheta2 = np.zeros([rd,rd])
dkdvar = np.zeros([rd,rd])
# dk dtheta for UU
UUdtheta1 = lambda dist: np.exp(-lu* dist)*dist + (-dist)*np.exp(-lu* dist)*(1+lu*dist)
UUdtheta2 = lambda dist: 0
UUdvar = lambda dist: (1 + lu *dist) * np.exp(-lu* dist)
# dk dtheta for YY
dk1theta1 = lambda dist: np.exp(-ly*dist)*2*(-lu)/(lu+ly)**3
#c=np.sqrt(3)
@ -216,7 +225,7 @@ class ODE_UY(Kernpart):
dk3theta1 = lambda dist: np.exp(-dist*lu)*(lu+ly)**(-2)*((2*lu+ly+dist*lu**2+lu*ly*dist)*(-dist-2/(lu+ly))+2+2*lu*dist+ly*dist)
dktheta1 = lambda dist: self.varianceU*self.varianceY*(dk1theta1+dk2theta1+dk3theta1)
#dktheta1 = lambda dist: self.varianceU*self.varianceY*(dk1theta1+dk2theta1+dk3theta1)
@ -230,14 +239,20 @@ class ODE_UY(Kernpart):
dk3theta2 = lambda dist: np.exp(-dist*lu) * (-3*lu-ly-dist*lu**2-lu*ly*dist)/(lu+ly)**3
dktheta2 = lambda dist: self.varianceU*self.varianceY*(dk1theta2 + dk2theta2 +dk3theta2)
#dktheta2 = lambda dist: self.varianceU*self.varianceY*(dk1theta2 + dk2theta2 +dk3theta2)
k1 = lambda dist: np.exp(-ly*dist)*(2*lu+ly)/(lu+ly)**2
k2 = lambda dist: (np.exp(-lu*dist)*(ly-2*lu+lu*ly*dist-lu**2*dist) + np.exp(-ly*dist)*(2*lu-ly) ) / (ly-lu)**2
k3 = lambda dist: np.exp(-lu*dist) * ( (1+lu*dist)/(lu+ly) + (lu)/(lu+ly)**2 )
dkdvar = k1+k2+k3
#dkdvar = k1+k2+k3
# dk dtheta for UY
for i, s1 in enumerate(slices):
@ -246,16 +261,35 @@ class ODE_UY(Kernpart):
for ss2 in s2:
if i==0 and j==0:
#target[ss1,ss2] = kuu(np.abs(rdist[ss1,ss2]))
#dktheta1[ss1,ss2] =
#dktheta2[ss1,ss2] =
#dkdvar[ss1,ss2] =
dktheta1[ss1,ss2] = self.varianceU*self.varianceY*UUdtheta1(rdist[ss1,ss2])
dktheta2[ss1,ss2] = 0
dkdvar[ss1,ss2] = self.varianceY*UUdvar(rdist[ss1,ss2])
elif i==0 and j==1:
#target[ss1,ss2] = np.where( rdist[ss1,ss2]>0 , kuyp(np.abs(rdist[ss1,ss2])), kuyn(np.abs(rdist[s1[0],s2[0]]) ) )
#dktheta1[ss1,ss2] =
#dktheta2[ss1,ss2] =
#dkdvar[ss1,ss2] =
dktheta1[ss1,ss2] = self.varianceU*self.varianceY*(dk1theta1(np.abs(rdist[ss1,ss2]))+dk2theta1(np.abs(rdist[ss1,ss2]))+dk3theta1(np.abs(rdist[ss1,ss2])))
dktheta2[ss1,ss2] = self.varianceU*self.varianceY*(dk1theta2(np.abs(rdist[ss1,ss2])) + dk2theta2(np.abs(rdist[ss1,ss2])) +dk3theta2(np.abs(rdist[ss1,ss2])))
dkdvar[ss1,ss2] = k1(np.abs(rdist[ss1,ss2]))+k2(np.abs(rdist[ss1,ss2]))+k3(np.abs(rdist[ss1,ss2]))
elif i==1 and j==1:
#target[ss1,ss2] = kyy(np.abs(rdist[ss1,ss2]))
dktheta1[ss1,ss2] = self.varianceU*self.varianceY*(dk1theta1(np.abs(rdist[ss1,ss2]))+dk2theta1(np.abs(rdist[ss1,ss2]))+dk3theta1(np.abs(rdist[ss1,ss2])))
dktheta2[ss1,ss2] = self.varianceU*self.varianceY*(dk1theta2(np.abs(rdist[ss1,ss2])) + dk2theta2(np.abs(rdist[ss1,ss2])) +dk3theta2(np.abs(rdist[ss1,ss2])))
dkdvar[ss1,ss2] = k1(np.abs(rdist[ss1,ss2]))+k2(np.abs(rdist[ss1,ss2]))+k3(np.abs(rdist[ss1,ss2]))
else:
#target[ss1,ss2] = np.where( rdist[ss1,ss2]>0 , kyup(np.abs(rdist[ss1,ss2])), kyun(np.abs(rdist[s1[0],s2[0]]) ) )
#dktheta1[ss1,ss2] =
#dktheta2[ss1,ss2] =
#dkdvar[ss1,ss2] =
dktheta1[ss1,ss2] = self.varianceU*self.varianceY*(dk1theta1(np.abs(rdist[ss1,ss2]))+dk2theta1(np.abs(rdist[ss1,ss2]))+dk3theta1(np.abs(rdist[ss1,ss2])))
dktheta2[ss1,ss2] = self.varianceU*self.varianceY*(dk1theta2(np.abs(rdist[ss1,ss2])) + dk2theta2(np.abs(rdist[ss1,ss2])) +dk3theta2(np.abs(rdist[ss1,ss2])))
dkdvar[ss1,ss2] = k1(np.abs(rdist[ss1,ss2]))+k2(np.abs(rdist[ss1,ss2]))+k3(np.abs(rdist[ss1,ss2]))
#stop
target[0] += np.sum(self.varianceY*dkdvar * dL_dK)