apunkt/GPy
1
0
Fork 0
mirror of https://github.com/SheffieldML/GPy.git synced 2026-05-09 12:02:38 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

merge for new kernel slice handling

Browse source
This commit is contained in:
Max Zwiessele 2014-03-14 10:56:51 +00:00
commit 2b9d3b326c
5 changed files with 32 additions and 34 deletions
Download patch file Download diff file Expand all files Collapse all files

11
GPy/core/model.py
View file

@ -15,7 +15,7 @@ import itertools
class Model(Parameterized): class Model(Parameterized):
_fail_count = 0 # Count of failed optimization steps (see objective) _fail_count = 0 # Count of failed optimization steps (see objective)
_allowed_failures = 10 # number of allowed failures _allowed_failures = 10 # number of allowed failures
def __init__(self, name): def __init__(self, name):
super(Model, self).__init__(name) # Parameterized.__init__(self) super(Model, self).__init__(name) # Parameterized.__init__(self)
self.optimization_runs = [] self.optimization_runs = []
@ -27,7 +27,7 @@ class Model(Parameterized):
def _log_likelihood_gradients(self): def _log_likelihood_gradients(self):
return self.gradient return self.gradient
def _getstate(self): def _getstate(self):
""" """
Get the current state of the class. Get the current state of the class.
@ -231,7 +231,7 @@ class Model(Parameterized):
raise RuntimeError, "Cannot optimize, when everything is fixed" raise RuntimeError, "Cannot optimize, when everything is fixed"
if self.size == 0: if self.size == 0:
raise RuntimeError, "Model without parameters cannot be minimized" raise RuntimeError, "Model without parameters cannot be minimized"
if optimizer is None: if optimizer is None:
optimizer = self.preferred_optimizer optimizer = self.preferred_optimizer
@ -301,9 +301,8 @@ class Model(Parameterized):
denominator = (2 * np.dot(dx, gradient)) denominator = (2 * np.dot(dx, gradient))
global_ratio = (f1 - f2) / np.where(denominator==0., 1e-32, denominator) global_ratio = (f1 - f2) / np.where(denominator==0., 1e-32, denominator)
gloabl_diff = (f1 - f2) - denominator
return (np.abs(1. - global_ratio) < tolerance) or (np.abs(gloabl_diff) == 0) return np.abs(1. - global_ratio) < tolerance
else: else:
# check the gradient of each parameter individually, and do some pretty printing # check the gradient of each parameter individually, and do some pretty printing
try: try:
@ -349,7 +348,7 @@ class Model(Parameterized):
xx[xind] -= 2.*step xx[xind] -= 2.*step
f2 = self.objective_function(xx) f2 = self.objective_function(xx)
numerical_gradient = (f1 - f2) / (2 * step) numerical_gradient = (f1 - f2) / (2 * step)
if np.all(gradient[xind]==0): ratio = (f1-f2) == gradient[xind] if np.all(gradient[xind]==0): ratio = (f1-f2) == gradient[xind]
else: ratio = (f1 - f2) / (2 * step * gradient[xind]) else: ratio = (f1 - f2) / (2 * step * gradient[xind])
difference = np.abs((f1 - f2) / 2 / step - gradient[xind]) difference = np.abs((f1 - f2) / 2 / step - gradient[xind])

2
GPy/core/sparse_gp.py
View file

@ -60,7 +60,7 @@ class SparseGP(GP):
dL_dKmm = self.grad_dict.pop('dL_dKmm') dL_dKmm = self.grad_dict.pop('dL_dKmm')
self.kern.update_gradients_full(dL_dKmm, self.Z, None) self.kern.update_gradients_full(dL_dKmm, self.Z, None)
target = self.kern.gradient.copy() target = self.kern.gradient.copy()
self.kern.update_gradients_expectations(variational_posterior=self.X, Z=self.Z, dL_dpsi0=grad_dict['dL_dpsi0'], dL_dpsi1=grad_dict['dL_dpsi1'], dL_dpsi2=grad_dict['dL_dpsi2']) self.kern.update_gradients_expectations(variational_posterior=self.X, Z=self.Z, dL_dpsi0=self.grad_dict['dL_dpsi0'], dL_dpsi1=self.grad_dict['dL_dpsi1'], dL_dpsi2=self.grad_dict['dL_dpsi2'])
self.kern.gradient += target self.kern.gradient += target
#gradients wrt Z #gradients wrt Z

2
GPy/kern/_src/independent_outputs.py
View file

@ -83,7 +83,7 @@ class IndependentOutputs(CombinationKernel):
target = np.zeros_like(X) target = np.zeros_like(X)
slices = index_to_slices(X[:,self.index_dim]) slices = index_to_slices(X[:,self.index_dim])
if X2 is None: if X2 is None:
[[np.copyto(target[s,self.kern.active_dims], self.kern.gradients_X(dL_dK[s,s],X[s],X[ss])) for s, ss in product(slices_i, slices_i)] for slices_i in slices] [[np.copyto(target[s,self.kern.active_dims], self.kern.gradients_X(dL_dK[s,ss],X[s],X[ss])) for s, ss in itertools.product(slices_i, slices_i)] for slices_i in slices]
else: else:
X2,slices2 = X2[:,:self.index_dim],index_to_slices(X2[:,-1]) X2,slices2 = X2[:,:self.index_dim],index_to_slices(X2[:,-1])
[[[np.copyto(target[s,:self.index_dim], self.kern.gradients_X(dL_dK[s,s2], X[s], X2[s2])) for s in slices_i] for s2 in slices_j] for slices_i,slices_j in zip(slices,slices2)] [[[np.copyto(target[s,:self.index_dim], self.kern.gradients_X(dL_dK[s,s2], X[s], X2[s2])) for s in slices_i] for s2 in slices_j] for slices_i,slices_j in zip(slices,slices2)]

49
GPy/testing/kernel_tests.py
View file

@ -252,34 +252,33 @@ class KernelGradientTestsContinuous(unittest.TestCase):
k.randomize() k.randomize()
self.assertTrue(check_kernel_gradient_functions(k, X=self.X, X2=self.X2, verbose=verbose)) self.assertTrue(check_kernel_gradient_functions(k, X=self.X, X2=self.X2, verbose=verbose))
#TODO: turn off grad checkingwrt X for indexed kernels liek coregionalize #TODO: turn off grad checkingwrt X for indexed kernels like coregionalize
class KernelGradientTestsContinuous1D(unittest.TestCase): # class KernelGradientTestsContinuous1D(unittest.TestCase):
def setUp(self): # def setUp(self):
self.N, self.D = 100, 1 # self.N, self.D = 100, 1
self.X = np.random.randn(self.N,self.D) # self.X = np.random.randn(self.N,self.D)
self.X2 = np.random.randn(self.N+10,self.D) # self.X2 = np.random.randn(self.N+10,self.D)
#
continuous_kerns = ['RBF', 'Linear'] # continuous_kerns = ['RBF', 'Linear']
self.kernclasses = [getattr(GPy.kern, s) for s in continuous_kerns] # self.kernclasses = [getattr(GPy.kern, s) for s in continuous_kerns]
#
def test_PeriodicExponential(self): # def test_PeriodicExponential(self):
k = GPy.kern.PeriodicExponential(self.D) # k = GPy.kern.PeriodicExponential(self.D)
k.randomize() # k.randomize()
self.assertTrue(check_kernel_gradient_functions(k, X=self.X, X2=self.X2, verbose=verbose)) # self.assertTrue(check_kernel_gradient_functions(k, X=self.X, X2=self.X2, verbose=verbose))
#
def test_PeriodicMatern32(self): # def test_PeriodicMatern32(self):
k = GPy.kern.PeriodicMatern32(self.D) # k = GPy.kern.PeriodicMatern32(self.D)
k.randomize() # k.randomize()
self.assertTrue(check_kernel_gradient_functions(k, X=self.X, X2=self.X2, verbose=verbose)) # self.assertTrue(check_kernel_gradient_functions(k, X=self.X, X2=self.X2, verbose=verbose))
#
def test_PeriodicMatern52(self): # def test_PeriodicMatern52(self):
k = GPy.kern.PeriodicMatern52(self.D) # k = GPy.kern.PeriodicMatern52(self.D)
k.randomize() # k.randomize()
self.assertTrue(check_kernel_gradient_functions(k, X=self.X, X2=self.X2, verbose=verbose)) # self.assertTrue(check_kernel_gradient_functions(k, X=self.X, X2=self.X2, verbose=verbose))
class KernelTestsMiscellaneous(unittest.TestCase): class KernelTestsMiscellaneous(unittest.TestCase):
def setUp(self): def setUp(self):
N, D = 100, 10 N, D = 100, 10
self.X = np.linspace(-np.pi, +np.pi, N)[:,None] * np.ones(D) self.X = np.linspace(-np.pi, +np.pi, N)[:,None] * np.ones(D)

2
GPy/util/caching.py
View file

@ -48,7 +48,7 @@ class Cacher(object):
if k in kw and kw[k] is not None: if k in kw and kw[k] is not None:
return self.operation(*args, **kw) return self.operation(*args, **kw)
# TODO: WARNING !!! Cache OFFSWITCH !!! WARNING # TODO: WARNING !!! Cache OFFSWITCH !!! WARNING
return self.operation(*args) #return self.operation(*args)
#if the result is cached, return the cached computation #if the result is cached, return the cached computation
state = [all(a is b for a, b in itertools.izip_longest(args, cached_i)) for cached_i in self.cached_inputs] state = [all(a is b for a, b in itertools.izip_longest(args, cached_i)) for cached_i in self.cached_inputs]