apunkt/GPy
1
0
Fork 0
mirror of https://github.com/SheffieldML/GPy.git synced 2026-05-09 03:52:39 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

non-working grads in linear

Browse source
This commit is contained in:
James Hensman 2014-02-21 11:25:33 +00:00
parent 81e0861a60
commit 0dc9a32ba3
3 changed files with 14 additions and 11 deletions
Download patch file Download diff file Expand all files Collapse all files

2
GPy/core/parameterization/param.py
View file

@ -147,7 +147,7 @@ class Param(Constrainable, ObservableArray, Gradcheckable, Indexable):
target += self.gradient.flat target += self.gradient.flat
def _set_gradient(self, g): def _set_gradient(self, g):
self.gradient = g self.gradient = g.reshape(self._realshape_)
#=========================================================================== #===========================================================================
# Array operations -> done # Array operations -> done

2
GPy/examples/dimensionality_reduction.py
View file

@ -25,7 +25,7 @@ def bgplvm_test_model(optimize=False, verbose=1, plot=False, output_dim=200, nan
#k = (GPy.kern.RBF(input_dim, .5, lengthscales, ARD=True) #k = (GPy.kern.RBF(input_dim, .5, lengthscales, ARD=True)
##+ GPy.kern.white(input_dim, 0.01) ##+ GPy.kern.white(input_dim, 0.01)
#) #)
k = GPy.kern.Linear(input_dim)# + GPy.kern.bias(input_dim) + GPy.kern.white(input_dim, 0.00001) k = GPy.kern.Linear(input_dim, ARD=1)# + GPy.kern.bias(input_dim) + GPy.kern.white(input_dim, 0.00001)
K = k.K(X) K = k.K(X)
Y = _np.random.multivariate_normal(_np.zeros(num_inputs), K, (output_dim,)).T Y = _np.random.multivariate_normal(_np.zeros(num_inputs), K, (output_dim,)).T

21
GPy/kern/_src/linear.py
View file

@ -47,7 +47,7 @@ class Linear(Kern):
self.variances = Param('variances', variances, Logexp()) self.variances = Param('variances', variances, Logexp())
self.add_parameter(self.variances) self.add_parameter(self.variances)
self.variances.add_observer(self._on_changed) self.variances.add_observer(self, self._on_changed)
def _on_changed(self, obj): def _on_changed(self, obj):
#TODO: move this to base class? isnt it jst for the caching? #TODO: move this to base class? isnt it jst for the caching?
@ -82,9 +82,9 @@ class Linear(Kern):
self._collect_gradient(target) self._collect_gradient(target)
self.update_gradients_full(dL_dKmm, Z, None) self.update_gradients_full(dL_dKmm, Z, None)
self._collect_gradient(target) self._collect_gradient(target)
return target self._set_gradient(target)
def update_gradients_full(self, dL_dK, X): def update_gradients_full(self, dL_dK, X, X2=None):
if self.ARD: if self.ARD:
if X2 is None: if X2 is None:
self.variances.gradient = np.array([np.sum(dL_dK * tdot(X[:, i:i + 1])) for i in range(self.input_dim)]) self.variances.gradient = np.array([np.sum(dL_dK * tdot(X[:, i:i + 1])) for i in range(self.input_dim)])
@ -130,16 +130,19 @@ class Linear(Kern):
def update_gradients_variational(self, dL_dKmm, dL_dpsi0, dL_dpsi1, dL_dpsi2, mu, S, Z): def update_gradients_variational(self, dL_dKmm, dL_dpsi0, dL_dpsi1, dL_dpsi2, mu, S, Z):
# psi0: # psi0:
tmp = dL_dpsi0[:, None] * self._mu2S(mu, S) tmp = dL_dpsi0[:, None] * self._mu2S(mu, S)
if self.ARD: self.variances.gradient[:] = tmp.sum(0) if self.ARD: grad = tmp.sum(0)
else: self.variances.gradient[:] = tmp.sum() else: grad = np.atleast_1d(tmp.sum())
#psi1 #psi1
self.variances.gradient += self._param_grad_helper(dL_dpsi1, mu, Z) self.update_gradients_full(dL_dpsi1, mu, Z)
grad += self.variances.gradient
#psi2 #psi2
tmp = dL_dpsi2[:, :, :, None] * (self._ZAinner(mu, S, Z)[:, :, None, :] * (2. * Z)[None, None, :, :]) tmp = dL_dpsi2[:, :, :, None] * (self._ZAinner(mu, S, Z)[:, :, None, :] * (2. * Z)[None, None, :, :])
if self.ARD: self.variances.gradient += tmp.sum(0).sum(0).sum(0) if self.ARD: grad += tmp.sum(0).sum(0).sum(0)
else: self.variances.gradient += tmp.sum() else: grad += tmp.sum()
#from Kmm #from Kmm
self.variances.gradient += self._param_grad_helper(dL_dKmm, Z, None) self.update_gradients_full(dL_dpsi1, mu, Z)
grad += self.variances.gradient
self._set_gradient(grad)
def gradients_Z_variational(self, dL_dKmm, dL_dpsi0, dL_dpsi1, dL_dpsi2, mu, S, Z): def gradients_Z_variational(self, dL_dKmm, dL_dpsi0, dL_dpsi1, dL_dpsi2, mu, S, Z):
# Kmm # Kmm