apunkt/GPy
1
0
Fork 0
mirror of https://github.com/SheffieldML/GPy.git synced 2026-04-27 13:56:23 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

weaved some slow functions in the stationary class. We now fall back (and latch) to numpy if weave fails

Browse source
This commit is contained in:
James Hensman 2014-10-16 01:28:41 +01:00
parent 5962035b54
commit f9a16059e4
1 changed files with 73 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

74
GPy/kern/_src/stationary.py
View file

@ -10,6 +10,7 @@ from ... import util
import numpy as np
from scipy import integrate
from ...util.caching import Cache_this
from ...util.config import config # for assesing whether to use weave
class Stationary(Kern):
"""
@ -139,7 +140,17 @@ class Stationary(Kern):
#self.lengthscale.gradient = -((dL_dr*rinv)[:,:,None]*x_xl3).sum(0).sum(0)/self.lengthscale**3
tmp = dL_dr*self._inv_dist(X, X2)
if X2 is None: X2 = X
self.lengthscale.gradient = np.array([np.einsum('ij,ij,...', tmp, np.square(X[:,q:q+1] - X2[:,q:q+1].T), -1./self.lengthscale[q]**3) for q in xrange(self.input_dim)])
if config.getboolean('weave', 'working'):
try:
self.lengthscale.gradient = self.weave_lengthscale_grads(tmp, X, X2)
except:
print "\n Weave compilation failed. Falling back to (slower) numpy implementation\n"
config.set('weave', 'working', 'False')
self.lengthscale.gradient = np.array([np.einsum('ij,ij,...', tmp, np.square(X[:,q:q+1] - X2[:,q:q+1].T), -1./self.lengthscale[q]**3) for q in xrange(self.input_dim)])
else:
self.lengthscale.gradient = np.array([np.einsum('ij,ij,...', tmp, np.square(X[:,q:q+1] - X2[:,q:q+1].T), -1./self.lengthscale[q]**3) for q in xrange(self.input_dim)])
else:
r = self._scaled_dist(X, X2)
self.lengthscale.gradient = -np.sum(dL_dr*r)/self.lengthscale
@ -154,10 +165,43 @@ class Stationary(Kern):
dist = self._scaled_dist(X, X2).copy()
return 1./np.where(dist != 0., dist, np.inf)
def weave_lengthscale_grads(self, tmp, X, X2):
N,M = tmp.shape
Q = X.shape[1]
if hasattr(X, 'values'):X = X.values
if hasattr(X2, 'values'):X2 = X2.values
grads = np.zeros(self.input_dim)
code = """
double gradq;
for(int q=0; q<Q; q++){
gradq = 0;
for(int n=0; n<N; n++){
for(int m=0; m<M; m++){
gradq += tmp(n,m)*(X(n,q)-X2(m,q))*(X(n,q)-X2(m,q));
}
}
grads[q] = gradq;
}
"""
from scipy import weave
weave.inline(code, ['tmp', 'X', 'X2', 'grads', 'N', 'M', 'Q'], type_converters=weave.converters.blitz, support_code="#include <math.h>")
return -grads/self.lengthscale**3
def gradients_X(self, dL_dK, X, X2=None):
"""
Given the derivative of the objective wrt K (dL_dK), compute the derivative wrt X
"""
if config.getboolean('weave', 'working'):
try:
return self.gradients_X_weave(dL_dK, X, X2)
except:
print "\n Weave compilation failed. Falling back to (slower) numpy implementation\n"
config.set('weave', 'working', 'False')
return self.gradients_X_(dL_dK, X, X2)
else:
return self.gradients_X_(dL_dK, X, X2)
def gradients_X_(self, dL_dK, X, X2=None):
invdist = self._inv_dist(X, X2)
dL_dr = self.dK_dr_via_X(X, X2) * dL_dK
tmp = invdist*dL_dr
@ -177,6 +221,34 @@ class Stationary(Kern):
return ret
def gradients_X_weave(self, dL_dK, X, X2=None):
invdist = self._inv_dist(X, X2)
dL_dr = self.dK_dr_via_X(X, X2) * dL_dK
tmp = invdist*dL_dr
if X2 is None:
tmp = tmp + tmp.T
X2 = X
ret = np.zeros(X.shape)
code = """
int n,q,d;
double retnd;
for(n=0;n<N;n++){
for(d=0;d<D;d++){
retnd = 0;
for(q=0;q<Q;q++){
retnd += tmp(n,q)*(X(n,d)-X2(q,d));
}
ret(n,d) = retnd;
}
}
"""
from scipy import weave
N,D = X.shape
Q = tmp.shape[1]
weave.inline(code, ['ret', 'N', 'D', 'Q', 'tmp', 'X', 'X2'], type_converters=weave.converters.blitz)
return ret/self.lengthscale**2
def gradients_X_diag(self, dL_dKdiag, X):
return np.zeros(X.shape)