apunkt/GPy
1
0
Fork 0
mirror of https://github.com/SheffieldML/GPy.git synced 2026-05-10 20:42:39 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

[gradients xx] getting there

Browse source
This commit is contained in:
Max Zwiessele 2016-05-04 09:11:04 +01:00
parent 4efb92c554
commit 5d19039d90
3 changed files with 52 additions and 51 deletions
Download patch file Download diff file Expand all files Collapse all files

2
GPy/core/gp.py
View file

@ -449,7 +449,7 @@ class GP(Model):
:param bool covariance: whether to include the covariance of the wishart embedding. :param bool covariance: whether to include the covariance of the wishart embedding.
:param array-like dimensions: which dimensions of the input space to use [defaults to self.get_most_significant_input_dimensions()[:2]] :param array-like dimensions: which dimensions of the input space to use [defaults to self.get_most_significant_input_dimensions()[:2]]
""" """
G = self.predict_wishard_embedding(Xnew, kern, mean, covariance) G = self.predict_wishart_embedding(Xnew, kern, mean, covariance)
if dimensions is None: if dimensions is None:
dimensions = self.get_most_significant_input_dimensions()[:2] dimensions = self.get_most_significant_input_dimensions()[:2]
G = G[:, dimensions][:,:,dimensions] G = G[:, dimensions][:,:,dimensions]

4
GPy/kern/src/kernel_slice_operations.py
View file

@ -70,11 +70,11 @@ class _Slice_wrap(object):
ret[:, self.k._all_dims_active] = return_val ret[:, self.k._all_dims_active] = return_val
elif len(self.shape) == 3: # derivative for X2!=None elif len(self.shape) == 3: # derivative for X2!=None
if self.diag: if self.diag:
ret[:, :, self.k._all_dims_active][:, self.k._all_dims_active] = return_val ret.T[np.ix_(self.k._all_dims_active, self.k._all_dims_active)] = return_val.T
else: else:
ret[:, :, self.k._all_dims_active] = return_val ret[:, :, self.k._all_dims_active] = return_val
elif len(self.shape) == 4: # second order derivative elif len(self.shape) == 4: # second order derivative
ret[:, :, self.k._all_dims_active][:, :, :, self.k._all_dims_active] = return_val ret.T[np.ix_(self.k._all_dims_active, self.k._all_dims_active)] = return_val.T
return ret return ret
return return_val return return_val

3
GPy/kern/src/stationary.py
View file

@ -254,8 +254,9 @@ class Stationary(Kern):
if cov: # full covariance if cov: # full covariance
grad = np.empty((X.shape[0], X2.shape[0], X2.shape[1], X.shape[1]), dtype=np.float64) grad = np.empty((X.shape[0], X2.shape[0], X2.shape[1], X.shape[1]), dtype=np.float64)
for q in range(self.input_dim): for q in range(self.input_dim):
tmpdist = (X[:,[q]]-X2[:,[q]].T)
for r in range(self.input_dim): for r in range(self.input_dim):
tmpdist2 = (X[:,[q]]-X2[:,[q]].T)*(X[:,[r]]-X2[:,[r]].T) # Introduce temporary distance tmpdist2 = tmpdist*(X[:,[r]]-X2[:,[r]].T) # Introduce temporary distance
if r==q: if r==q:
grad[:, :, q, r] = np.multiply(dL_dK,(np.multiply((tmp1*invdist2 - tmp2),tmpdist2)/l2[r] - tmp1)/l2[q]) grad[:, :, q, r] = np.multiply(dL_dK,(np.multiply((tmp1*invdist2 - tmp2),tmpdist2)/l2[r] - tmp1)/l2[q])
else: else: