def make_input(self, batch_dict, train=False):
"""Transform a `batch_dict` into a TensorFlow `feed_dict`
:param batch_dict: (``dict``) A dictionary containing all inputs to the embeddings for this model
:param train: (``bool``) Are we training. Defaults to False
:return:
"""
y = batch_dict.get('y', None)
if not tf.executing_eagerly():
batch_for_model = new_placeholder_dict(train)
for k in self.embeddings.keys():
batch_for_model["{}:0".format(k)] = batch_dict[k]
# Allow us to track a length, which is needed for BLSTMs
if self.lengths_key is not None:
batch_for_model[self.lengths] = batch_dict[self.lengths_key]
if y is not None:
batch_for_model[self.y] = fill_y(len(self.labels), y)
else:
SET_TRAIN_FLAG(train)
batch_for_model = {}
for k in self.embeddings.keys():
batch_for_model[k] = batch_dict[k]
# Allow us to track a length, which is needed for BLSTMs
if self.lengths_key is not None:
batch_for_model["lengths"] = batch_dict[self.lengths_key]
return batch_for_model
def ex2dict(self, x, y, do_dropout=False):
pkeep = 1.0 - self.pdrop_value if do_dropout else 1
return {
self.x: x,
self.y: fill_y(len(self.labels), y),
self.pkeep: pkeep
}
def make_input(self, batch_dict, do_dropout=False):
x = batch_dict['x']
y = batch_dict['y']
lengths = batch_dict['lengths']
pkeep = 1.0 - self.pdrop_value if do_dropout else 1
return {
self.x: x,
self.lengths: lengths,
self.y: fill_y(len(self.labels), y),
self.pkeep: pkeep
}
def ex2dict(self, x, y, do_dropout=False):
"""Convert from an input of x and y tensors to a `feed_dict`
:param x: Input tensor `x` (`BxT`)
:param y: Input tensor `y` (`B`)
:param do_dropout: Defaults to off. If its on, use the dropout value provided during model construction
:return: A `feed_dict`
"""
pkeep = 1.0 - self.pdrop_value if do_dropout else 1
return {
self.x: x,
self.y: fill_y(len(self.labels), y),
self.pkeep: pkeep
}
def make_input(self, batch_dict, do_dropout=False):
x = batch_dict['x']
y = batch_dict.get('y', None)
xch = batch_dict.get('xch')
lengths = batch_dict.get('lengths')
pkeep = 1.0 - self.pdrop_value if do_dropout else 1.0
feed_dict = {self.x: x, self.pkeep: pkeep}
if hasattr(self, 'lengths') and self.lengths is not None:
feed_dict[self.lengths] = lengths
if hasattr(self, 'xch') and xch is not None and self.xch is not None:
feed_dict[self.xch] = xch
if y is not None:
feed_dict[self.y] = fill_y(len(self.labels), y)
return feed_dict
def make_input(self, batch_dict, train=False):
if train is False:
return self.inference.make_input(batch_dict)
y = batch_dict.get('y', None)
feed_dict = new_placeholder_dict(True)
for key in self.parallel_params.keys():
feed_dict["{}_parallel:0".format(key)] = batch_dict[key]
# Allow us to track a length, which is needed for BLSTMs
if self.lengths_key is not None:
feed_dict[self.lengths] = batch_dict[self.lengths_key]
if y is not None:
feed_dict[self.y] = fill_y(len(self.labels), y)
return feed_dict
def make_input(self, batch_dict, train=False):
if train is False:
return self.inference.make_input(batch_dict)
y = batch_dict.get('y', None)
feed_dict = new_placeholder_dict(True)
for key in self.parallel_params.keys():
feed_dict["{}_parallel:0".format(key)] = batch_dict[key]
# Allow us to track a length, which is needed for BLSTMs
if self.lengths_key is not None:
feed_dict[self.lengths] = batch_dict[self.lengths_key]
if y is not None:
feed_dict[self.y] = fill_y(len(self.labels), y)
return feed_dict
def _test(self, ts):
total_loss = 0
steps = len(ts)
cm = ConfusionMatrix(self.model.labels)
pg = create_progress_bar(steps)
for batch_dict in ts:
y = fill_y(len(self.model.labels), batch_dict['y'])
feed_dict = self.model.make_input(batch_dict, do_dropout=True)
preds, lossv, = self.model.sess.run(
[self.model.best, self.model.loss], feed_dict=feed_dict)
# print(preds)
cm.add_batch(y, preds)
total_loss += lossv
pg.update()
pg.done()
metrics = cm.get_all_metrics()
metrics['avg_loss'] = float(total_loss) / steps
return metrics
def make_input(self, batch_dict, train=False):
"""Transform a `batch_dict` into a TensorFlow `feed_dict`
:param batch_dict: (``dict``) A dictionary containing all inputs to the embeddings for this model
:param train: (``bool``) Are we training. Defaults to False
:return:
"""
y = batch_dict.get('y', None)
feed_dict = new_placeholder_dict(train)
for k in self.embeddings.keys():
feed_dict["{}:0".format(k)] = batch_dict[k]
# Allow us to track a length, which is needed for BLSTMs
if self.lengths_key is not None:
feed_dict[self.lengths] = batch_dict[self.lengths_key]
if y is not None:
feed_dict[self.y] = fill_y(len(self.labels), y)
return feed_dict
