def testGetLossRaisesErrors(self):
with self.assertRaisesRegex(
TypeError,
'custom_name.*must be.*MinDiffLoss.*string.*4.*int'):
utils._get_loss(4, 'custom_name')
with self.assertRaisesRegex(
ValueError,
'custom_name.*must be.*supported values.*bad_name'):
utils._get_loss('bad_name', 'custom_name')
def testForMMDLoss(self):
loss = utils._get_loss('mmd')
self.assertIsInstance(loss, mmd_loss.MMDLoss)
loss = utils._get_loss('mmd_loss')
self.assertIsInstance(loss, mmd_loss.MMDLoss)
loss = utils._get_loss(mmd_loss.MMDLoss())
self.assertIsInstance(loss, mmd_loss.MMDLoss)
loss_name = 'custom_name'
loss = utils._get_loss(mmd_loss.MMDLoss(name=loss_name))
self.assertIsInstance(loss, mmd_loss.MMDLoss)
self.assertEqual(loss.name, loss_name)
def __init__(self,
original_model: tf.keras.Model,
loss,
loss_weight: complex = 1.0,
predictions_transform=None,
**kwargs):
"""Initializes a MinDiffModel instance.
Raises:
ValueError: If `predictions_transform` is passed in but not callable.
"""
super(MinDiffModel, self).__init__(**kwargs)
# Set _auto_track_sub_layers to true to ensure we track the
# original_model and MinDiff layers.
self._auto_track_sub_layers = True # Track sub layers.
self.built = True # This Model is built, original_model may or may not be.
self._original_model = original_model
self._loss = loss_utils._get_loss(loss)
self._loss_weight = loss_weight
self._min_diff_loss_metric = tf.keras.metrics.Mean("min_diff_loss")
if (predictions_transform is not None
and not callable(predictions_transform)):
raise ValueError(
"`predictions_transform` must be callable if passed "
"in, given: {}".format(predictions_transform))
self._predictions_transform = predictions_transform
# Clear input_spec in case there is one. We cannot make any strong
# assertions because `min_diff_data` may or may not be included and can
# have different shapes since weight is optional.
self.input_spec = None
def testForCustomLoss(self):
class CustomLoss(base_loss.MinDiffLoss):
def call(self, x, y):
pass
loss = CustomLoss()
loss_output = utils._get_loss(loss)
self.assertIs(loss_output, loss)
def testForAbsoluteCorrelationLoss(self):
loss = utils._get_loss('abs_corr')
self.assertIsInstance(loss, abscorrloss.AbsoluteCorrelationLoss)
loss = utils._get_loss('abS_coRr') # Strangely capitalized.
self.assertIsInstance(loss, abscorrloss.AbsoluteCorrelationLoss)
loss = utils._get_loss('abs_corr_loss') # Other accepted name.
self.assertIsInstance(loss, abscorrloss.AbsoluteCorrelationLoss)
loss = utils._get_loss('absolute_correlation') # Other accepted name.
self.assertIsInstance(loss, abscorrloss.AbsoluteCorrelationLoss)
loss = utils._get_loss(
'absolute_correlation_loss') # Other accepted name.
self.assertIsInstance(loss, abscorrloss.AbsoluteCorrelationLoss)
loss_name = 'custom_name'
loss = utils._get_loss(abscorrloss.AbsoluteCorrelationLoss(loss_name))
self.assertIsInstance(loss, abscorrloss.AbsoluteCorrelationLoss)
self.assertEqual(loss.name, loss_name)
def testAcceptsNone(self):
loss = utils._get_loss(None)
self.assertIsNone(loss)
