mirror of
https://github.com/SheffieldML/GPy.git
synced 2026-05-09 20:12:38 +02:00
kern merge commencing
This commit is contained in:
Max Zwiessele
parent
1294c24a28
commit
f8ff2c7df2
1 changed files with 111 additions and 120 deletions
|
|
@ -5,130 +5,121 @@ Created on 11 Mar 2014
|
||||||
'''
|
'''
|
||||||
from ...core.parameterization.parameterized import ParametersChangedMeta
|
from ...core.parameterization.parameterized import ParametersChangedMeta
|
||||||
import numpy as np
|
import numpy as np
|
||||||
|
import functools
|
||||||
|
|
||||||
class KernCallsViaSlicerMeta(ParametersChangedMeta):
|
class KernCallsViaSlicerMeta(ParametersChangedMeta):
|
||||||
def __call__(self, *args, **kw):
|
def __call__(self, *args, **kw):
|
||||||
instance = super(ParametersChangedMeta, self).__call__(*args, **kw)
|
instance = super(ParametersChangedMeta, self).__call__(*args, **kw)
|
||||||
instance.K = _slice_wrapper(instance, instance.K)
|
instance.K = _Slice_wrapper(instance, instance.K)
|
||||||
instance.Kdiag = _slice_wrapper(instance, instance.Kdiag, diag=True)
|
instance.Kdiag = _Slice_wrapper_diag(instance, instance.Kdiag)
|
||||||
instance.update_gradients_full = _slice_wrapper(instance, instance.update_gradients_full, diag=False, derivative=True)
|
|
||||||
instance.update_gradients_diag = _slice_wrapper(instance, instance.update_gradients_diag, diag=True, derivative=True)
|
instance.update_gradients_full = _Slice_wrapper_derivative(instance, instance.update_gradients_full)
|
||||||
instance.gradients_X = _slice_wrapper(instance, instance.gradients_X, diag=False, derivative=True, ret_X=True)
|
instance.update_gradients_diag = _Slice_wrapper_diag_derivative(instance, instance.update_gradients_diag)
|
||||||
instance.gradients_X_diag = _slice_wrapper(instance, instance.gradients_X_diag, diag=True, derivative=True, ret_X=True)
|
|
||||||
instance.psi0 = _slice_wrapper(instance, instance.psi0, diag=False, derivative=False)
|
instance.gradients_X = _Slice_wrapper_grad_X(instance, instance.gradients_X)
|
||||||
instance.psi1 = _slice_wrapper(instance, instance.psi1, diag=False, derivative=False)
|
instance.gradients_X_diag = _Slice_wrapper_grad_X_diag(instance, instance.gradients_X_diag)
|
||||||
instance.psi2 = _slice_wrapper(instance, instance.psi2, diag=False, derivative=False)
|
|
||||||
instance.update_gradients_expectations = _slice_wrapper(instance, instance.update_gradients_expectations, derivative=True, psi_stat=True)
|
instance.psi0 = _Slice_wrapper(instance, instance.psi0)
|
||||||
instance.gradients_Z_expectations = _slice_wrapper(instance, instance.gradients_Z_expectations, derivative=True, psi_stat_Z=True, ret_X=True)
|
instance.psi1 = _Slice_wrapper(instance, instance.psi1)
|
||||||
instance.gradients_qX_expectations = _slice_wrapper(instance, instance.gradients_qX_expectations, derivative=True, psi_stat=True, ret_X=True)
|
instance.psi2 = _Slice_wrapper(instance, instance.psi2)
|
||||||
|
|
||||||
|
instance.update_gradients_expectations = _Slice_wrapper_psi_stat_derivative_no_ret(instance, instance.update_gradients_expectations)
|
||||||
|
instance.gradients_Z_expectations = _Slice_wrapper_psi_stat_derivative_Z(instance, instance.gradients_Z_expectations)
|
||||||
|
instance.gradients_qX_expectations = _Slice_wrapper_psi_stat_derivative(instance, instance.gradients_qX_expectations)
|
||||||
instance.parameters_changed()
|
instance.parameters_changed()
|
||||||
return instance
|
return instance
|
||||||
|
|
||||||
def _slice_wrapper(kern, operation, diag=False, derivative=False, psi_stat=False, psi_stat_Z=False, ret_X=False):
|
class _Slice_wrap(object):
|
||||||
"""
|
def __init__(self, instance, f):
|
||||||
This method wraps the functions in kernel to make sure all kernels allways see their respective input dimension.
|
self.k = instance
|
||||||
The different switches are:
|
self.f = f
|
||||||
diag: if X2 exists
|
def copy_to(self, new_instance):
|
||||||
derivative: if first arg is dL_dK
|
return self.__class__(new_instance, self.f)
|
||||||
psi_stat: if first 3 args are dL_dpsi0..2
|
def _slice_X(self, X):
|
||||||
psi_stat_Z: if first 2 args are dL_dpsi1..2
|
return self.k._slice_X(X) if not self.k._sliced_X else X
|
||||||
"""
|
def _slice_X_X2(self, X, X2):
|
||||||
if derivative:
|
return self.k._slice_X(X) if not self.k._sliced_X else X, self.k._slice_X(X2) if X2 is not None and not self.k._sliced_X else X2
|
||||||
if diag:
|
def __enter__(self):
|
||||||
def x_slice_wrapper(dL_dKdiag, X):
|
self.k._sliced_X += 1
|
||||||
ret_X_not_sliced = ret_X and kern._sliced_X == 0
|
return self
|
||||||
if ret_X_not_sliced:
|
def __exit__(self, *a):
|
||||||
ret = np.zeros(X.shape)
|
self.k._sliced_X -= 1
|
||||||
X = kern._slice_X(X) if not kern._sliced_X else X
|
|
||||||
# if the return value is of shape X.shape, we need to make sure to return the right shape
|
class _Slice_wrapper(_Slice_wrap):
|
||||||
kern._sliced_X += 1
|
def __call__(self, X, X2 = None, *a, **kw):
|
||||||
try:
|
X, X2 = self._slice_X_X2(X, X2)
|
||||||
if ret_X_not_sliced: ret[:, kern.active_dims] = operation(dL_dKdiag, X)
|
with self:
|
||||||
else: ret = operation(dL_dKdiag, X)
|
ret = self.f(X, X2, *a, **kw)
|
||||||
except:
|
return ret
|
||||||
raise
|
|
||||||
finally:
|
class _Slice_wrapper_diag(_Slice_wrap):
|
||||||
kern._sliced_X -= 1
|
def __call__(self, X, *a, **kw):
|
||||||
return ret
|
X = self._slice_X(X)
|
||||||
elif psi_stat:
|
with self:
|
||||||
def x_slice_wrapper(dL_dpsi0, dL_dpsi1, dL_dpsi2, Z, variational_posterior):
|
ret = self.f(X, *a, **kw)
|
||||||
ret_X_not_sliced = ret_X and kern._sliced_X == 0
|
return ret
|
||||||
if ret_X_not_sliced:
|
|
||||||
ret1, ret2 = np.zeros(variational_posterior.shape), np.zeros(variational_posterior.shape)
|
class _Slice_wrapper_derivative(_Slice_wrap):
|
||||||
Z, variational_posterior = kern._slice_X(Z) if not kern._sliced_X else Z, kern._slice_X(variational_posterior) if not kern._sliced_X else variational_posterior
|
def __call__(self, dL_dK, X, X2=None):
|
||||||
kern._sliced_X += 1
|
self._slice_X(X)
|
||||||
# if the return value is of shape X.shape, we need to make sure to return the right shape
|
with self:
|
||||||
try:
|
ret = self.f(dL_dK, X, X2)
|
||||||
if ret_X_not_sliced:
|
return ret
|
||||||
ret = list(operation(dL_dpsi0, dL_dpsi1, dL_dpsi2, Z, variational_posterior))
|
|
||||||
r2 = ret[:2]
|
class _Slice_wrapper_diag_derivative(_Slice_wrap):
|
||||||
ret[0] = ret1
|
def __call__(self, dL_dKdiag, X):
|
||||||
ret[1] = ret2
|
X = self._slice_X(X)
|
||||||
ret[0][:, kern.active_dims] = r2[0]
|
with self:
|
||||||
ret[1][:, kern.active_dims] = r2[1]
|
ret = self.f(dL_dKdiag, X)
|
||||||
del r2
|
return ret
|
||||||
else: ret = operation(dL_dpsi0, dL_dpsi1, dL_dpsi2, Z, variational_posterior)
|
|
||||||
except:
|
class _Slice_wrapper_grad_X(_Slice_wrap):
|
||||||
raise
|
def __call__(self, dL_dK, X, X2=None):
|
||||||
finally:
|
ret = np.zeros(X.shape)
|
||||||
kern._sliced_X -= 1
|
X, X2 = self._slice_X_X2(X, X2)
|
||||||
return ret
|
with self:
|
||||||
elif psi_stat_Z:
|
ret[:, self.k.active_dims] = self.f(dL_dK, X, X2)
|
||||||
def x_slice_wrapper(dL_dpsi1, dL_dpsi2, Z, variational_posterior):
|
return ret
|
||||||
ret_X_not_sliced = ret_X and kern._sliced_X == 0
|
|
||||||
if ret_X_not_sliced: ret = np.zeros(Z.shape)
|
class _Slice_wrapper_grad_X_diag(_Slice_wrap):
|
||||||
Z, variational_posterior = kern._slice_X(Z) if not kern._sliced_X else Z, kern._slice_X(variational_posterior) if not kern._sliced_X else variational_posterior
|
def __call__(self, dL_dKdiag, X):
|
||||||
kern._sliced_X += 1
|
ret = np.zeros(X.shape)
|
||||||
try:
|
X = self._slice_X(X)
|
||||||
if ret_X_not_sliced:
|
with self:
|
||||||
ret[:, kern.active_dims] = operation(dL_dpsi1, dL_dpsi2, Z, variational_posterior)
|
ret[:, self.k.active_dims] = self.f(dL_dKdiag, X)
|
||||||
else: ret = operation(dL_dpsi1, dL_dpsi2, Z, variational_posterior)
|
return ret
|
||||||
except:
|
|
||||||
raise
|
class _Slice_wrapper_psi_stat_derivative_no_ret(_Slice_wrap):
|
||||||
finally:
|
def __call__(self, dL_dpsi0, dL_dpsi1, dL_dpsi2, Z, variational_posterior):
|
||||||
kern._sliced_X -= 1
|
Z, variational_posterior = self._slice_X_X2(Z, variational_posterior)
|
||||||
return ret
|
with self:
|
||||||
else:
|
ret = self.f(dL_dpsi0, dL_dpsi1, dL_dpsi2, Z, variational_posterior)
|
||||||
def x_slice_wrapper(dL_dK, X, X2=None):
|
return ret
|
||||||
ret_X_not_sliced = ret_X and kern._sliced_X == 0
|
|
||||||
if ret_X_not_sliced:
|
class _Slice_wrapper_psi_stat_derivative(_Slice_wrap):
|
||||||
ret = np.zeros(X.shape)
|
def __call__(self, dL_dpsi0, dL_dpsi1, dL_dpsi2, Z, variational_posterior):
|
||||||
X, X2 = kern._slice_X(X) if not kern._sliced_X else X, kern._slice_X(X2) if X2 is not None and not kern._sliced_X else X2
|
ret1, ret2 = np.zeros(variational_posterior.shape), np.zeros(variational_posterior.shape)
|
||||||
kern._sliced_X += 1
|
Z, variational_posterior = self._slice_X_X2(Z, variational_posterior)
|
||||||
try:
|
with self:
|
||||||
if ret_X_not_sliced: ret[:, kern.active_dims] = operation(dL_dK, X, X2)
|
ret = list(self.f(dL_dpsi0, dL_dpsi1, dL_dpsi2, Z, variational_posterior))
|
||||||
else: ret = operation(dL_dK, X, X2)
|
r2 = ret[:2]
|
||||||
except:
|
ret[0] = ret1
|
||||||
raise
|
ret[1] = ret2
|
||||||
finally:
|
ret[0][:, self.k.active_dims] = r2[0]
|
||||||
kern._sliced_X -= 1
|
ret[1][:, self.k.active_dims] = r2[1]
|
||||||
return ret
|
del r2
|
||||||
else:
|
return ret
|
||||||
if diag:
|
|
||||||
def x_slice_wrapper(X, *args, **kw):
|
class _Slice_wrapper_psi_stat_derivative_Z(_Slice_wrap):
|
||||||
X = kern._slice_X(X) if not kern._sliced_X else X
|
def __call__(self, dL_dpsi0, dL_dpsi1, dL_dpsi2, Z, variational_posterior):
|
||||||
kern._sliced_X += 1
|
ret1, ret2 = np.zeros(variational_posterior.shape), np.zeros(variational_posterior.shape)
|
||||||
try:
|
Z, variational_posterior = self._slice_X_X2(Z, variational_posterior)
|
||||||
ret = operation(X, *args, **kw)
|
with self:
|
||||||
except:
|
ret = list(self.f(dL_dpsi0, dL_dpsi1, dL_dpsi2, Z, variational_posterior))
|
||||||
raise
|
r2 = ret[:2]
|
||||||
finally:
|
ret[0] = ret1
|
||||||
kern._sliced_X -= 1
|
ret[1] = ret2
|
||||||
return ret
|
ret[0][:, self.k.active_dims] = r2[0]
|
||||||
else:
|
ret[1][:, self.k.active_dims] = r2[1]
|
||||||
def x_slice_wrapper(X, X2=None, *args, **kw):
|
del r2
|
||||||
X, X2 = kern._slice_X(X) if not kern._sliced_X else X, kern._slice_X(X2) if X2 is not None and not kern._sliced_X else X2
|
return ret
|
||||||
kern._sliced_X += 1
|
|
||||||
try:
|
|
||||||
ret = operation(X, X2, *args, **kw)
|
|
||||||
except: raise
|
|
||||||
finally:
|
|
||||||
kern._sliced_X -= 1
|
|
||||||
return ret
|
|
||||||
x_slice_wrapper._operation = operation
|
|
||||||
x_slice_wrapper.__name__ = ("slicer("+str(operation)
|
|
||||||
+(","+str(bool(diag)) if diag else'')
|
|
||||||
+(','+str(bool(derivative)) if derivative else '')
|
|
||||||
+')')
|
|
||||||
x_slice_wrapper.__doc__ = "**sliced**\n" + (operation.__doc__ or "")
|
|
||||||
return x_slice_wrapper
|
|
||||||
|
|
|
||||||
Loading…
Add table
Add a link
Reference in a new issue