https://github.com/GPflow/GPflow
Tip revision: 03af631bdbefbe67f85c81e74df10b860d05588f authored by Artem Artemev on 03 November 2019, 22:18:47 UTC
Resolve merge conflicts from develop
Resolve merge conflicts from develop
Tip revision: 03af631
scipy_optimizer.py
# Copyright 2017 Artem Artemev @awav
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from ..core.compilable import Build
from ..core.errors import GPflowError
from ..models.model import Model
from . import external_optimizer, optimizer
from tensorflow.python.framework.errors_impl import InvalidArgumentError
class ScipyOptimizer(optimizer.Optimizer):
def __init__(self, **kwargs):
self._optimizer_kwargs = kwargs
self._optimizer = None
self._model = None
def make_optimize_tensor(self, model, session=None, var_list=None, **kwargs):
"""
Make SciPy optimization tensor.
The `make_optimize_tensor` method builds optimization tensor and initializes
all necessary variables created by optimizer.
:param model: GPflow model.
:param session: Tensorflow session.
:param var_list: List of variables for training.
:param kwargs: Scipy optional optimization parameters,
- `maxiter`, maximal number of iterations to perform.
- `disp`, if True, prints convergence messages.
:return: Tensorflow operation.
"""
session = model.enquire_session(session)
with session.as_default():
var_list = self._gen_var_list(model, var_list)
optimizer_kwargs = self._optimizer_kwargs.copy()
options = optimizer_kwargs.get('options', {})
options.update(kwargs)
optimizer_kwargs.update(dict(options=options))
objective = model.objective
optimizer = external_optimizer.ScipyOptimizerInterface(objective, var_list=var_list, **optimizer_kwargs)
return optimizer
def minimize(self,
model,
session=None,
var_list=None,
feed_dict=None,
maxiter=1000,
disp=False,
initialize=False,
initialize_optimizer=False,
anchor=True,
step_callback=None,
**kwargs):
"""
Minimizes objective function of the model.
:param model: GPflow model with objective tensor.
:param session: Session where optimization will be run.
:param var_list: List of extra variables which should be trained during optimization.
:param feed_dict: Feed dictionary of tensors passed to session run method.
:param maxiter: Number of run interation. Note: scipy optimizer can do early stopping
if model converged.
:param disp: ScipyOptimizer option. Set to True to print convergence messages.
:param initialize: If `True` model parameters will be re-initialized even if they were
initialized before for gotten session.
:param initialize_optimizer: If `True` model parameters will be re-initialized even if they were
initialized before for gotten session.
:param anchor: If `True` trained parameters computed during optimization at
particular session will be synchronized with internal parameter values.
:param step_callback: A function to be called at each optimization step;
arguments are the current values of all optimization variables
flattened into a single vector.
:type step_callback: Callable[[np.ndarray], None]
:param kwargs: This is a dictionary of extra parameters for session run method.
"""
if model is None or not isinstance(model, Model):
raise ValueError('Unknown type passed for optimization.')
if model.is_built_coherence() is Build.NO:
raise GPflowError('Model is not built.')
if self._model is not None and self._model is not model:
raise ValueError("Optimizer used with another model. Create new optimizer or reset existing.")
existing_optimizer = self._optimizer is not None
if not existing_optimizer:
self._optimizer = self.make_optimize_tensor(model, session, var_list=var_list, maxiter=maxiter, disp=disp)
if self._model is None or initialize:
model.initialize(session=session)
if self._model is None:
self._model = model
feed_dict = self._gen_feed_dict(model, feed_dict)
session = model.enquire_session(session)
if existing_optimizer and not initialize_optimizer:
try:
options = dict(options=dict(maxiter=maxiter, disp=disp))
self._optimizer.optimizer_kwargs.update(options)
self._optimizer.optimize(session=session, feed_dict=feed_dict, step_callback=step_callback)
except InvalidArgumentError as error:
msg = ("This error might occur because the internal state "
"(for example, variables shape or dtype) of the model "
"is changed. "
"In this case, you have to use a new optimiser")
msg = "Original error message: \n\t{error}\nOptimiser message: {msg}.".format(
error=error, msg=msg)
raise RuntimeError(msg)
except Exception as error:
msg = ("Unknown error has occured at reusage of the "
"ScipyOptimizer. Make sure that you use the same "
"TensorFlow session.")
msg = "Original error message: \n\t{error}\nOptimiser message: {msg}.".format(
error=error, msg=msg)
raise RuntimeError(msg)
else:
self._optimizer.minimize(session=session, feed_dict=feed_dict, step_callback=step_callback, **kwargs)
if anchor:
model.anchor(session)
@property
def model(self):
return self._model
@property
def optimizer(self):
return self._optimizer