def build(self, optimizer=None, metric='accuracy', **kwargs):
Predictor.build(self,
optimizer=optimizer,
loss='cross_entropy',
metric=metric,
metric_is_like_loss=False,
metric_name='Accuracy')
def _print_progress(self, epc, start_time, info_dict, **kwargs):
# Sanity check
assert self._metric is not None
if self._validation_set is None:
raise ValueError('Validation set is None')
# Reset dict
info_dict = {}
data_batch = kwargs.get('data_batch', None)
assert data_batch is not None
feed_dict = self._get_default_feed_dict(data_batch, is_training=False)
sum_train_acc, train_acc = self._session.run(
[self._sum_train_acc, self._metric], feed_dict=feed_dict)
feed_dict = self._get_default_feed_dict(self._validation_set,
is_training=False)
sum_val_acc, val_acc = self._session.run(
[self._sum_val_acc, self._metric], feed_dict=feed_dict)
self._summary_writer.add_summary(sum_train_acc, self._counter)
self._summary_writer.add_summary(sum_val_acc, self._counter)
info_dict['Training Accuracy'] = train_acc
info_dict['Validation Accuracy'] = val_acc
# Call predictor's _print_progress
Predictor._print_progress(self, epc, start_time, info_dict)
def __init__(self, mark=None, net_type=Feedforward):
Predictor.__init__(self, mark, net_type)
# Private attributes
self._probabilities = TensorSlot(self, 'Probability')
self._evaluation_group = Group(self,
self._metric,
self._probabilities,
name='evaluation group')
def _build(self, optimizer=None, metric=None, **kwargs):
# TODO: ... do some compromise
hub.block_validation = True
# If last layer is not softmax layer, add it to model
if not (isinstance(self.last_function, Activation)
and self.last_function.abbreviation == 'softmax'):
self.add(Activation('softmax'))
# Call parent's method to build using the default loss function
# -- cross entropy
Predictor._build(self, optimize=optimizer, metric=metric, **kwargs)
assert self.outputs.activated
# Plug tensor into probabilities slot
self._probabilities.plug(self.outputs.tensor)
def build(self,
optimizer=None,
loss='cross_entropy',
metric='accuracy',
batch_metric=None,
eval_metric=None,
**kwargs):
Predictor.build(self,
optimizer=optimizer,
loss=loss,
metric=metric,
batch_metric=batch_metric,
eval_metric=eval_metric,
**kwargs)
def build(self, loss='cross_entropy', optimizer=None):
# TODO: ... do some compromise
config.block_validation = True
# Call parent's method to build using the default loss function
# -- cross entropy
Predictor.build(self,
loss,
optimizer,
metric='accuracy',
metric_name=pedia.Accuracy)
self._sum_train_acc = tf.summary.scalar('train_acc', self._metric)
self._sum_val_acc = tf.summary.scalar('val_acc', self._metric)
# Find last layer
if (isinstance(self.last_function, Activation)
and self.last_function.abbreviation == 'softmax'):
self._probabilities = self.outputs
else:
self._probabilities = tf.nn.softmax(self.outputs,
name='probabilities')
def __init__(self, mark=None, net_type=Feedforward):
Predictor.__init__(self, mark, net_type)
# Private attributes
self._probabilities = TensorSlot(self, 'Probability')
def __init__(self, mark=None):
Predictor.__init__(self, mark)
self._sum_train_acc = None
self._sum_val_acc = None
self._probabilities = None