apunkt/GPy
1
0
Fork 0
mirror of https://github.com/SheffieldML/GPy.git synced 2026-06-05 14:55:15 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

working on coregionalize

Browse source
This commit is contained in:
James Hensman 2014-02-21 17:32:40 +00:00
parent 78ceef01c3
commit fddc663f28
1 changed files with 44 additions and 47 deletions
Download patch file Download diff file Expand all files Collapse all files

59
GPy/kern/_src/coregionalize.py
View file

@ -5,6 +5,7 @@ from kern import Kern
import numpy as np import numpy as np
from scipy import weave from scipy import weave
from ...core.parameterization import Param from ...core.parameterization import Param
from ...core.parameterization.transformations import Logexp
class Coregionalize(Kern): class Coregionalize(Kern):
""" """
@ -37,7 +38,7 @@ class Coregionalize(Kern):
super(Coregionalize, self).__init__(input_dim=1, name=name) super(Coregionalize, self).__init__(input_dim=1, name=name)
self.output_dim = output_dim self.output_dim = output_dim
self.rank = rank self.rank = rank
if self.rank>output_dim-1: if self.rank>output_dim:
print("Warning: Unusual choice of rank, it should normally be less than the output_dim.") print("Warning: Unusual choice of rank, it should normally be less than the output_dim.")
if W is None: if W is None:
W = 0.5*np.random.randn(self.output_dim, self.rank)/np.sqrt(self.rank) W = 0.5*np.random.randn(self.output_dim, self.rank)/np.sqrt(self.rank)
@ -48,7 +49,7 @@ class Coregionalize(Kern):
kappa = 0.5*np.ones(self.output_dim) kappa = 0.5*np.ones(self.output_dim)
else: else:
assert kappa.shape==(self.output_dim, ) assert kappa.shape==(self.output_dim, )
self.kappa = Param('kappa', kappa) self.kappa = Param('kappa', kappa, Logexp())
self.add_parameters(self.W, self.kappa) self.add_parameters(self.W, self.kappa)
self.parameters_changed() self.parameters_changed()
@ -56,8 +57,8 @@ class Coregionalize(Kern):
def parameters_changed(self): def parameters_changed(self):
self.B = np.dot(self.W, self.W.T) + np.diag(self.kappa) self.B = np.dot(self.W, self.W.T) + np.diag(self.kappa)
def K(self,index,index2,target): def K(self, X, X2=None):
index = np.asarray(index,dtype=np.int) index = np.asarray(X, dtype=np.int)
#here's the old code (numpy) #here's the old code (numpy)
#if index2 is None: #if index2 is None:
@ -69,41 +70,45 @@ class Coregionalize(Kern):
#ii, jj = ii.T, jj.T #ii, jj = ii.T, jj.T
#false_target += self.B[ii, jj] #false_target += self.B[ii, jj]
if index2 is None:
if X2 is None:
target = np.empty((X.shape[0], X.shape[0]), dtype=np.float64)
code=""" code="""
for(int i=0;i<N; i++){ for(int i=0;i<N; i++){
target[i+i*N] += B[index[i]+output_dim*index[i]]; target[i+i*N] = B[index[i]+output_dim*index[i]];
for(int j=0; j<i; j++){ for(int j=0; j<i; j++){
target[j+i*N] += B[index[i]+output_dim*index[j]]; target[j+i*N] = B[index[i]+output_dim*index[j]];
target[i+j*N] += target[j+i*N]; target[i+j*N] = target[j+i*N];
} }
} }
""" """
N, B, output_dim = index.size, self.B, self.output_dim N, B, output_dim = index.size, self.B, self.output_dim
weave.inline(code, ['target', 'index', 'N', 'B', 'output_dim']) weave.inline(code, ['target', 'index', 'N', 'B', 'output_dim'])
else: else:
index2 = np.asarray(index2,dtype=np.int) index2 = np.asarray(X2, dtype=np.int)
target = np.empty((X.shape[0], X2.shape[0]), dtype=np.float64)
code=""" code="""
for(int i=0;i<num_inducing; i++){ for(int i=0;i<num_inducing; i++){
for(int j=0; j<N; j++){ for(int j=0; j<N; j++){
target[i+j*num_inducing] += B[output_dim*index[j]+index2[i]]; target[i+j*num_inducing] = B[output_dim*index[j]+index2[i]];
} }
} }
""" """
N, num_inducing, B, output_dim = index.size, index2.size, self.B, self.output_dim N, num_inducing, B, output_dim = index.size, index2.size, self.B, self.output_dim
weave.inline(code, ['target', 'index', 'index2', 'N', 'num_inducing', 'B', 'output_dim']) weave.inline(code, ['target', 'index', 'index2', 'N', 'num_inducing', 'B', 'output_dim'])
return target
def Kdiag(self,index,target): def Kdiag(self, X):
target += np.diag(self.B)[np.asarray(index,dtype=np.int).flatten()] return np.diag(self.B)[np.asarray(X, dtype=np.int).flatten()]
def update_gradients_full(self,dL_dK, index, index2=None): def update_gradients_full(self, dL_dK, X, X2=None):
index = np.asarray(index,dtype=np.int) index = np.asarray(X, dtype=np.int)
dL_dK_small = np.zeros_like(self.B) dL_dK_small = np.zeros_like(self.B)
if index2 is None: if X2 is None:
index2 = index index2 = index
else: else:
index2 = np.asarray(index2,dtype=np.int) index2 = np.asarray(X2, dtype=np.int)
code=""" code="""
for(int i=0; i<num_inducing; i++){ for(int i=0; i<num_inducing; i++){
@ -122,19 +127,11 @@ class Coregionalize(Kern):
self.W.gradient = dW self.W.gradient = dW
self.kappa.gradient = dkappa self.kappa.gradient = dkappa
def update_gradients_sparse(self, dL_dKmm, dL_dKnm, dL_dKdiag, X, Z): def update_gradients_diag(self, dL_dKdiag, X):
raise NotImplementedError, "some code below" index = np.asarray(X, dtype=np.int).flatten()
#def dKdiag_dtheta(self,dL_dKdiag,index,target): dL_dKdiag_small = np.array([dL_dKdiag[index==i] for i in xrange(output_dim)])
#index = np.asarray(index,dtype=np.int).flatten() self.W.gradient = 2.*self.W*dL_dKdiag_small[:, None]
#dL_dKdiag_small = np.zeros(self.output_dim) self.kappa.gradient = dL_dKdiag_small
#for i in range(self.output_dim):
#dL_dKdiag_small[i] += np.sum(dL_dKdiag[index==i])
#dW = 2.*self.W*dL_dKdiag_small[:,None]
#dkappa = dL_dKdiag_small
#target += np.hstack([dW.flatten(),dkappa])
def gradients_X(self,dL_dK,X,X2): def gradients_X(self, dL_dK, X, X2=None):
if X2 is None: return np.zeros(X.shape)
return np.zeros((X.shape[0], X.shape[0]))
else:
return np.zeros((X.shape[0], X2.shape[0]))