https://github.com/GPflow/GPflow
Tip revision: 6fd1a26809a4c754b73c1a645b48d7cda35b2cd6 authored by John Bradshaw on 24 October 2017, 10:29:09 UTC
Merge remote-tracking branch 'origin/GPflow-1.0-RC' into john-bradshaw/linear-features-for-kernels-gpflow1.0
Merge remote-tracking branch 'origin/GPflow-1.0-RC' into john-bradshaw/linear-features-for-kernels-gpflow1.0
Tip revision: 6fd1a26
decors.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.
import functools
import contextlib
import tensorflow as tf
from .core.base import GPflowError
from .core.base import Build
from .core.node import Node
from .core.autoflow import AutoFlow
from .core.autoflow import TemplateFlow
from .core.tensor_converter import TensorConverter
from .params import Parameterized
def name_scope(name=None):
def name_scope_wrapper(method):
@functools.wraps(method)
def runnable(*args, **kwargs):
scope_name = name if name is not None else method.__name__
with tf.name_scope(scope_name):
return method(*args, **kwargs)
return runnable
return name_scope_wrapper
def params_as_tensors(method):
@functools.wraps(method)
def tensor_mode_wrapper(obj, *args, **kwargs):
if not isinstance(obj, Parameterized):
raise GPflowError(
'Tensor mode works only for parmeterized object.')
prev_value = _params_as_tensors_enter(obj, True)
try:
result = method(obj, *args, **kwargs)
finally:
_params_as_tensors_exit(obj, prev_value)
return result
return tensor_mode_wrapper
@contextlib.contextmanager
def params_as_tensors_for(obj, convert=True):
prev_value = _params_as_tensors_enter(obj, convert)
try:
yield
finally:
_params_as_tensors_exit(obj, prev_value)
def templateflow(template_name_for_tf, create_scope_now_=False, unique_name_=None,
custom_getter_=None, **kwargs_template):
def templateflow_wrapper(method_to_be_wrapped):
@functools.wraps(method_to_be_wrapped)
def runnable(obj, *args, **kwargs):
if not isinstance(obj, Node):
raise GPflowError('Templateflow mode works only for node-like object.')
graph = obj.enquire_graph()
if obj.is_built(graph) is not Build.YES:
raise ValueError("Trying to use a templated function before the node upon which it"
" is defined on has been built.")
name = method_to_be_wrapped.__name__
store = TemplateFlow.get_autoflow(obj, name)
template = store.pop("template", None)
if template is None:
scope_name = _name_scope_name(obj, name)
with graph.as_default(), tf.name_scope(scope_name):
template_name = template_name_for_tf
method = functools.partial(method_to_be_wrapped, obj)
template = tf.make_template(template_name, method,
create_scope_now_=create_scope_now_,
unique_name_=unique_name_,
custom_getter_=custom_getter_, **kwargs_template)
store["template"] = template
return template(*args, **kwargs)
return runnable
return templateflow_wrapper
def autoflow(*af_args, **af_kwargs):
def autoflow_wrapper(method):
@functools.wraps(method)
def runnable(obj, *args, **kwargs):
if not isinstance(obj, Node):
raise GPflowError(
'Tensor mode works only for node-like object.')
if obj.is_built_coherence(obj.graph) is Build.NO:
raise GPflowError('Compilable object is not built.')
name = method.__name__
store = AutoFlow.get_autoflow(obj, name)
session = kwargs.pop('session', None)
session = obj.enquire_session(session=session)
if not store:
scope_name = _name_scope_name(obj, name)
with session.graph.as_default(), tf.name_scope(scope_name):
_setup_storage(store, *af_args, **af_kwargs)
previous_unitialised_vars = _get_set_of_unit_var_names(session)
_build_method(method, obj, store)
current_unitialised_vars = _get_set_of_unit_var_names(session)
session.run(tf.variables_initializer(
_collect_vars_with_name(current_unitialised_vars - previous_unitialised_vars)
))
return _session_run(session, obj, store, *args, **kwargs)
return runnable
return autoflow_wrapper
def _collect_vars_with_name(names):
return [v for v in tf.global_variables() if v.name.split(':')[0]
in names]
def _get_set_of_unit_var_names(session):
return set(_convert_bytes_to_strings(
session.run(tf.report_uninitialized_variables(tf.global_variables()))))
def _convert_bytes_to_strings(byte_iter):
return (x.decode("utf-8") for x in byte_iter)
def _params_as_tensors_enter(obj, convert=True):
name = TensorConverter.__tensor_mode__
attr_value = getattr(obj, name, None)
setattr(obj, name, convert)
return attr_value
def _params_as_tensors_exit(obj, previous):
name = TensorConverter.__tensor_mode__
if previous is not None:
setattr(obj, name, previous)
else:
delattr(obj, name)
def _setup_storage(store, *args, **_kwargs):
store['arguments'] = [tf.placeholder(*arg) for arg in args]
def _name_scope_name(obj, name):
return '/'.join(['autoflow', obj.name, name])
def _session_run(session, obj, store, *args, **kwargs):
feed_dict_key = 'feed_dict'
if feed_dict_key not in kwargs:
kwargs[feed_dict_key] = {}
feed_dict = kwargs.get(feed_dict_key)
feed_dict.update(dict(zip(store['arguments'], args)))
if obj.feeds:
feed_dict.update(obj.feeds)
return session.run(store['result'], **kwargs)
def _build_method(method, obj, store):
store['result'] = method(obj, *store['arguments'])