########################################################################################
#
# Hierarchical Attentive Recurrent Tracking
# Copyright (C) 2017 Adam R. Kosiorek, Oxford Robotics Institute, University of Oxford
# email: adamk@robots.ox.ac.uk
# webpage: http://ori.ox.ac.uk
#
# This program is free software; you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation; either version 3 of the License, or
# (at your option) any later version.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU General Public License for more details.
# You should have received a copy of the GNU General Public License
# along with this program. If not, see <http://www.gnu.org/licenses/>.
#
########################################################################################
import tensorflow as tf
from neurocity.component.model import base
class Model(base.BaseModel):
"""An abstraction of a Neural Network model. It simplifies layer management and supports
train/test modes. To define a new model, derive a class and overload the _build method,
which should be responsible for constructing the model, e.g.:
class MLP(object):
def _build(self):
self.inpt = tf.placeholder(tf.float32, (32, 100), name='inpt')
l1 = AffineLayer(self.inpt, 200)
l2 = AffineLayer(l1, 10)
Both layers will be registered in model.layers attribute.
It is important to call:
model.train_mode() - before training
model.test_mode() - before testing
Note: `mode` can be overwritten by calling one of the global setters:
neurocity.train_mode() or neurocity.test_mode()"""
def __init__(self, name='Model'):
super(Model, self).__init__()
self.name = name
self.layers = []
model_vars = set(tf.get_collection(tf.GraphKeys.MODEL_VARIABLES))
trainable_vars = set(tf.trainable_variables())
with tf.variable_scope(self.name):
with self:
self._build()
base.get_model().register(self)
self.trainable_vars = set(tf.trainable_variables()) - trainable_vars
self.model_vars = set(tf.get_collection(tf.GraphKeys.MODEL_VARIABLES)) - model_vars - self.trainable_vars
@property
def vars(self):
return self.trainable_vars.union(self.model_vars)
def saver(self, **kwargs):
"""Returns a Saver for all (trainable and model) variables used by the model.
Model variables include e.g. moving mean and average in BatchNorm.
:return: tf.Saver
"""
return tf.train.Saver(self.vars, **kwargs)
def _build(self):
raise NotImplementedError
def __enter__(self):
base.set_model(self)
return self
def __exit__(self, exc_type, exc_val, exc_tb):
base.reset_model()
if exc_type is None:
return True
