apunkt/GPy
1
0
Fork 0
mirror of https://github.com/SheffieldML/GPy.git synced 2026-05-18 13:55:14 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

removed ipdb statement from kern, cleaned up some nasty whitespace

Browse source
This commit is contained in:
James Hensman 2013-11-27 13:16:18 +00:00
parent 6673a8ae02
commit db9e5314e4
1 changed files with 21 additions and 23 deletions
Download patch file Download diff file Expand all files Collapse all files

44
GPy/kern/kern.py
View file

@ -487,12 +487,11 @@ class kern(Parameterized):
p1.psi1(Z, mu, S, psi11) p1.psi1(Z, mu, S, psi11)
Mu, Sigma = p1._crossterm_mu_S(Z, mu, S) Mu, Sigma = p1._crossterm_mu_S(Z, mu, S)
Mu, Sigma = Mu.reshape(NM,self.input_dim), Sigma.reshape(NM,self.input_dim) Mu, Sigma = Mu.reshape(NM,self.input_dim), Sigma.reshape(NM,self.input_dim)
p2.psi1(Z, Mu, Sigma, psi12) p2.psi1(Z, Mu, Sigma, psi12)
eK2 = psi12.reshape(N, M, M) eK2 = psi12.reshape(N, M, M)
crossterms = eK2 * (psi11[:, :, None] + psi11[:, None, :]) crossterms = eK2 * (psi11[:, :, None] + psi11[:, None, :])
target += crossterms target += crossterms
#import ipdb;ipdb.set_trace()
else: else:
raise NotImplementedError, "psi2 cannot be computed for this kernel" raise NotImplementedError, "psi2 cannot be computed for this kernel"
return target return target
@ -540,15 +539,15 @@ class kern(Parameterized):
# turn around to have rbf in front # turn around to have rbf in front
p1, p2 = self.parts[i2], self.parts[i1] p1, p2 = self.parts[i2], self.parts[i1]
ps1, ps2 = self.param_slices[i2], self.param_slices[i1] ps1, ps2 = self.param_slices[i2], self.param_slices[i1]
N, M = mu.shape[0], Z.shape[0]; NM=N*M N, M = mu.shape[0], Z.shape[0]; NM=N*M
psi11 = np.zeros((N, M)) psi11 = np.zeros((N, M))
p1.psi1(Z, mu, S, psi11) p1.psi1(Z, mu, S, psi11)
Mu, Sigma = p1._crossterm_mu_S(Z, mu, S) Mu, Sigma = p1._crossterm_mu_S(Z, mu, S)
Mu, Sigma = Mu.reshape(NM,self.input_dim), Sigma.reshape(NM,self.input_dim) Mu, Sigma = Mu.reshape(NM,self.input_dim), Sigma.reshape(NM,self.input_dim)
tmp1 = np.zeros_like(target[ps1]) tmp1 = np.zeros_like(target[ps1])
tmp2 = np.zeros_like(target[ps2]) tmp2 = np.zeros_like(target[ps2])
# for n in range(N): # for n in range(N):
@ -559,7 +558,7 @@ class kern(Parameterized):
# Mu, Sigma= Mu.reshape(N,M,self.input_dim), Sigma.reshape(N,M,self.input_dim) # Mu, Sigma= Mu.reshape(N,M,self.input_dim), Sigma.reshape(N,M,self.input_dim)
# p2.dpsi1_dtheta((dL_dpsi2[n:n+1,m:m+1,m_prime:m_prime+1]*(psi11[n:n+1,m_prime:m_prime+1]))[0], Z[m:m+1], Mu[n:n+1,m], Sigma[n:n+1,m], target[ps2]) # p2.dpsi1_dtheta((dL_dpsi2[n:n+1,m:m+1,m_prime:m_prime+1]*(psi11[n:n+1,m_prime:m_prime+1]))[0], Z[m:m+1], Mu[n:n+1,m], Sigma[n:n+1,m], target[ps2])
# p2.dpsi1_dtheta((dL_dpsi2[n:n+1,m:m+1,m_prime:m_prime+1]*(psi11[n:n+1,m:m+1]))[0], Z[m_prime:m_prime+1], Mu[n:n+1, m_prime], Sigma[n:n+1, m_prime], target[ps2])#Z[m_prime:m_prime+1], Mu[n+m:(n+m)+1], Sigma[n+m:(n+m)+1], target[ps2]) # p2.dpsi1_dtheta((dL_dpsi2[n:n+1,m:m+1,m_prime:m_prime+1]*(psi11[n:n+1,m:m+1]))[0], Z[m_prime:m_prime+1], Mu[n:n+1, m_prime], Sigma[n:n+1, m_prime], target[ps2])#Z[m_prime:m_prime+1], Mu[n+m:(n+m)+1], Sigma[n+m:(n+m)+1], target[ps2])
if isinstance(p1, RBF) and isinstance(p2, RBF): if isinstance(p1, RBF) and isinstance(p2, RBF):
psi12 = np.zeros((N, M)) psi12 = np.zeros((N, M))
p2.psi1(Z, mu, S, psi12) p2.psi1(Z, mu, S, psi12)
@ -571,11 +570,11 @@ class kern(Parameterized):
if isinstance(p1, RBF) and isinstance(p2, Linear): if isinstance(p1, RBF) and isinstance(p2, Linear):
#import ipdb;ipdb.set_trace() #import ipdb;ipdb.set_trace()
pass pass
p2.dpsi1_dtheta((dL_dpsi2*(psi11[:,:,None] + psi11[:,None,:])).reshape(NM,M), Z, Mu, Sigma, tmp2) p2.dpsi1_dtheta((dL_dpsi2*(psi11[:,:,None] + psi11[:,None,:])).reshape(NM,M), Z, Mu, Sigma, tmp2)
target[ps1] += tmp1 target[ps1] += tmp1
target[ps2] += tmp2 target[ps2] += tmp2
else: else:
raise NotImplementedError, "psi2 cannot be computed for this kernel" raise NotImplementedError, "psi2 cannot be computed for this kernel"
@ -615,17 +614,17 @@ class kern(Parameterized):
psi11 = np.zeros((N, M)) psi11 = np.zeros((N, M))
psi12 = np.zeros((NM, M)) psi12 = np.zeros((NM, M))
#psi12_t = np.zeros((N,M)) #psi12_t = np.zeros((N,M))
p1.psi1(Z, mu, S, psi11) p1.psi1(Z, mu, S, psi11)
Mu, Sigma = p1._crossterm_mu_S(Z, mu, S) Mu, Sigma = p1._crossterm_mu_S(Z, mu, S)
Mu, Sigma = Mu.reshape(NM,self.input_dim), Sigma.reshape(NM,self.input_dim) Mu, Sigma = Mu.reshape(NM,self.input_dim), Sigma.reshape(NM,self.input_dim)
p2.psi1(Z, Mu, Sigma, psi12) p2.psi1(Z, Mu, Sigma, psi12)
tmp1 = np.zeros_like(target) tmp1 = np.zeros_like(target)
p1.dpsi1_dZ((dL_dpsi2*psi12.reshape(N,M,M)).sum(1), Z, mu, S, tmp1) p1.dpsi1_dZ((dL_dpsi2*psi12.reshape(N,M,M)).sum(1), Z, mu, S, tmp1)
p1.dpsi1_dZ((dL_dpsi2*psi12.reshape(N,M,M)).sum(2), Z, mu, S, tmp1) p1.dpsi1_dZ((dL_dpsi2*psi12.reshape(N,M,M)).sum(2), Z, mu, S, tmp1)
target += tmp1 target += tmp1
#p2.dpsi1_dtheta((dL_dpsi2*(psi11[:,:,None] + psi11[:,None,:])).reshape(NM,M), Z, Mu, Sigma, target) #p2.dpsi1_dtheta((dL_dpsi2*(psi11[:,:,None] + psi11[:,None,:])).reshape(NM,M), Z, Mu, Sigma, target)
p2.dpsi1_dZ((dL_dpsi2*(psi11[:,:,None] + psi11[:,None,:])).reshape(NM,M), Z, Mu, Sigma, target) p2.dpsi1_dZ((dL_dpsi2*(psi11[:,:,None] + psi11[:,None,:])).reshape(NM,M), Z, Mu, Sigma, target)
else: else:
@ -666,21 +665,21 @@ class kern(Parameterized):
psi11 = np.zeros((N, M)) psi11 = np.zeros((N, M))
psi12 = np.zeros((NM, M)) psi12 = np.zeros((NM, M))
#psi12_t = np.zeros((N,M)) #psi12_t = np.zeros((N,M))
p1.psi1(Z, mu, S, psi11) p1.psi1(Z, mu, S, psi11)
Mu, Sigma = p1._crossterm_mu_S(Z, mu, S) Mu, Sigma = p1._crossterm_mu_S(Z, mu, S)
Mu, Sigma = Mu.reshape(NM,self.input_dim), Sigma.reshape(NM,self.input_dim) Mu, Sigma = Mu.reshape(NM,self.input_dim), Sigma.reshape(NM,self.input_dim)
p2.psi1(Z, Mu, Sigma, psi12) p2.psi1(Z, Mu, Sigma, psi12)
p1.dpsi1_dmuS((dL_dpsi2*psi12.reshape(N,M,M)).sum(1), Z, mu, S, target_mu, target_S) p1.dpsi1_dmuS((dL_dpsi2*psi12.reshape(N,M,M)).sum(1), Z, mu, S, target_mu, target_S)
p1.dpsi1_dmuS((dL_dpsi2*psi12.reshape(N,M,M)).sum(2), Z, mu, S, target_mu, target_S) p1.dpsi1_dmuS((dL_dpsi2*psi12.reshape(N,M,M)).sum(2), Z, mu, S, target_mu, target_S)
#p2.dpsi1_dtheta((dL_dpsi2*(psi11[:,:,None] + psi11[:,None,:])).reshape(NM,M), Z, Mu, Sigma, target) #p2.dpsi1_dtheta((dL_dpsi2*(psi11[:,:,None] + psi11[:,None,:])).reshape(NM,M), Z, Mu, Sigma, target)
p2.dpsi1_dmuS((dL_dpsi2*(psi11[:,:,None])).sum(1)*2, Z, Mu.reshape(N,M,self.input_dim).sum(1), Sigma.reshape(N,M,self.input_dim).sum(1), target_mu, target_S) p2.dpsi1_dmuS((dL_dpsi2*(psi11[:,:,None])).sum(1)*2, Z, Mu.reshape(N,M,self.input_dim).sum(1), Sigma.reshape(N,M,self.input_dim).sum(1), target_mu, target_S)
else: else:
raise NotImplementedError, "psi2 cannot be computed for this kernel" raise NotImplementedError, "psi2 cannot be computed for this kernel"
return target_mu, target_S return target_mu, target_S
def plot(self, x=None, plot_limits=None, which_parts='all', resolution=None, *args, **kwargs): def plot(self, x=None, plot_limits=None, which_parts='all', resolution=None, *args, **kwargs):
if which_parts == 'all': if which_parts == 'all':
which_parts = [True] * self.num_parts which_parts = [True] * self.num_parts
@ -753,7 +752,7 @@ class Kern_check_model(Model):
dL_dK = np.ones((X.shape[0], X.shape[0])) dL_dK = np.ones((X.shape[0], X.shape[0]))
else: else:
dL_dK = np.ones((X.shape[0], X2.shape[0])) dL_dK = np.ones((X.shape[0], X2.shape[0]))
self.kernel=kernel self.kernel=kernel
self.X = X self.X = X
self.X2 = X2 self.X2 = X2
@ -768,7 +767,7 @@ class Kern_check_model(Model):
return False return False
else: else:
return True return True
def _get_params(self): def _get_params(self):
return self.kernel._get_params() return self.kernel._get_params()
@ -783,7 +782,7 @@ class Kern_check_model(Model):
def _log_likelihood_gradients(self): def _log_likelihood_gradients(self):
raise NotImplementedError, "This needs to be implemented to use the kern_check_model class." raise NotImplementedError, "This needs to be implemented to use the kern_check_model class."
class Kern_check_dK_dtheta(Kern_check_model): class Kern_check_dK_dtheta(Kern_check_model):
"""This class allows gradient checks for the gradient of a kernel with respect to parameters. """ """This class allows gradient checks for the gradient of a kernel with respect to parameters. """
def __init__(self, kernel=None, dL_dK=None, X=None, X2=None): def __init__(self, kernel=None, dL_dK=None, X=None, X2=None):
@ -798,7 +797,7 @@ class Kern_check_dKdiag_dtheta(Kern_check_model):
Kern_check_model.__init__(self,kernel=kernel,dL_dK=dL_dK, X=X, X2=None) Kern_check_model.__init__(self,kernel=kernel,dL_dK=dL_dK, X=X, X2=None)
if dL_dK==None: if dL_dK==None:
self.dL_dK = np.ones((self.X.shape[0])) self.dL_dK = np.ones((self.X.shape[0]))
def log_likelihood(self): def log_likelihood(self):
return (self.dL_dK*self.kernel.Kdiag(self.X)).sum() return (self.dL_dK*self.kernel.Kdiag(self.X)).sum()
@ -815,7 +814,7 @@ class Kern_check_dK_dX(Kern_check_model):
def _get_param_names(self): def _get_param_names(self):
return ['X_' +str(i) + ','+str(j) for j in range(self.X.shape[1]) for i in range(self.X.shape[0])] return ['X_' +str(i) + ','+str(j) for j in range(self.X.shape[1]) for i in range(self.X.shape[0])]
def _get_params(self): def _get_params(self):
return self.X.flatten() return self.X.flatten()
@ -837,7 +836,7 @@ class Kern_check_dKdiag_dX(Kern_check_model):
def _get_param_names(self): def _get_param_names(self):
return ['X_' +str(i) + ','+str(j) for j in range(self.X.shape[1]) for i in range(self.X.shape[0])] return ['X_' +str(i) + ','+str(j) for j in range(self.X.shape[1]) for i in range(self.X.shape[0])]
def _get_params(self): def _get_params(self):
return self.X.flatten() return self.X.flatten()
@ -863,7 +862,6 @@ def kern_test(kern, X=None, X2=None, output_ind=None, verbose=False, X_positive=
if output_ind is not None: if output_ind is not None:
assert(output_ind<kern.input_dim) assert(output_ind<kern.input_dim)
X[:, output_ind] = np.random.randint(low=0,high=kern.parts[0].output_dim, size=X.shape[0]) X[:, output_ind] = np.random.randint(low=0,high=kern.parts[0].output_dim, size=X.shape[0])
import ipdb; ipdb.set_trace()
if X2==None: if X2==None:
X2 = np.random.randn(20, kern.input_dim) X2 = np.random.randn(20, kern.input_dim)
if X_positive: if X_positive: