def categorical_crossentropy(expected, predicted):
"""
Categorical cross-entropy error.
Parameters
----------
{error_function.expected}
{error_function.predicted}
Returns
-------
{error_function.Returns}
"""
epsilon = smallest_positive_number()
predicted = T.clip(predicted, epsilon, 1.0 - epsilon)
return T.nnet.categorical_crossentropy(predicted, expected).mean()
def categorical_crossentropy(expected, predicted):
"""
Categorical cross-entropy error.
Parameters
----------
{error_function.expected}
{error_function.predicted}
Returns
-------
{error_function.Returns}
"""
epsilon = smallest_positive_number()
predicted = T.clip(predicted, epsilon, 1.0 - epsilon)
return T.nnet.categorical_crossentropy(predicted, expected).mean()
def test_smallest_positive_number(self):
epsilon = smallest_positive_number()
self.assertNotEqual(0, asfloat(1) - (asfloat(1) - asfloat(epsilon)))
self.assertEqual(0, asfloat(1) - (asfloat(1) - asfloat(epsilon / 10)))
def categorical_crossentropy(expected, predicted):
""" Categorical cross-entropy error.
"""
epsilon = smallest_positive_number()
predicted = T.clip(predicted, epsilon, 1.0 - epsilon)
return T.nnet.categorical_crossentropy(predicted, expected).mean()
def binary_crossentropy(expected, predicted):
""" Binary cross-entropy error.
"""
epsilon = smallest_positive_number()
predicted = T.clip(predicted, epsilon, 1.0 - epsilon)
return T.nnet.binary_crossentropy(predicted, expected).mean()
def test_smallest_positive_number(self):
epsilon = smallest_positive_number()
self.assertNotEqual(0, asfloat(1) - (asfloat(1) - asfloat(epsilon)))
self.assertEqual(0, asfloat(1) - (asfloat(1) - asfloat(epsilon / 10)))
def categorical_crossentropy(expected, predicted):
""" Categorical cross-entropy error.
"""
epsilon = smallest_positive_number()
predicted = T.clip(predicted, epsilon, 1.0 - epsilon)
return T.nnet.categorical_crossentropy(predicted, expected).mean()
def binary_crossentropy(expected, predicted):
""" Binary cross-entropy error.
"""
epsilon = smallest_positive_number()
predicted = T.clip(predicted, epsilon, 1.0 - epsilon)
return T.nnet.binary_crossentropy(predicted, expected).mean()