apunkt/GPy
1
0
Fork 0
mirror of https://github.com/SheffieldML/GPy.git synced 2026-05-24 14:15:14 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

robust failure handling in model objective and gradient

Browse source
This commit is contained in:
Max Zwiessele 2013-06-07 14:07:21 +01:00
parent f2eb611e6a
commit 73bed5be82
2 changed files with 23 additions and 5 deletions
Download patch file Download diff file Expand all files Collapse all files

26
GPy/core/model.py
View file

@ -14,6 +14,8 @@ from numpy.linalg.linalg import LinAlgError
# import numdifftools as ndt
class Model(Parameterised):
_fail_count = 0 # Count of failed optimization steps (see objective)
_allowed_failures = 10 # number of allowed failures
def __init__(self):
Parameterised.__init__(self)
self.priors = None
@ -227,23 +229,36 @@ class Model(Parameterised):
def objective_function(self, x):
"""
The objective function passed to the optimizer. It combines the likelihood and the priors.
Failures are handled robustly. The algorithm will try several times to
return the objective, and will raise the original exception if it
the objective cannot be computed.
"""
try:
self._set_params_transformed(x)
self._fail_count = 0
except (LinAlgError, ZeroDivisionError) as e:
print "DEBUG: catching linalg, rejecting step" # TODO: delete
if self._fail_count >= self._allowed_failures:
raise e
self._fail_count += 1
return np.inf
return -self.log_likelihood() - self.log_prior()
def objective_function_gradients(self, x):
"""
Gets the gradients from the likelihood and the priors.
Failures are handled robustly. The algorithm will try several times to
return the objective, and will raise the original exception if it
the objective cannot be computed.
"""
try:
self._set_params_transformed(x)
self._fail_count = 0
except (LinAlgError, ZeroDivisionError) as e:
print "DEBUG: catching linalg, rejecting step" # TODO: delete
# return np.ones_like(x)
if self._fail_count >= self._allowed_failures:
raise e
self._fail_count += 1
obj_grads = -self._transform_gradients(self._log_likelihood_gradients() + self._log_prior_gradients())
return obj_grads
@ -251,8 +266,11 @@ class Model(Parameterised):
try:
self._set_params_transformed(x)
obj_f = -self.log_likelihood() - self.log_prior()
self._fail_count = 0
except (LinAlgError, ZeroDivisionError) as e:
print "DEBUG: catching linalg, rejecting step" # TODO: delete
if self._fail_count >= self._allowed_failures:
raise e
self._fail_count += 1
obj_f = np.inf
obj_grads = -self._transform_gradients(self._log_likelihood_gradients() + self._log_prior_gradients())
return obj_f, obj_grads