def test_cross_entropy_matrix():
assert error.CrossEntropyError()(numpy.array([[0.1, 1.], [1., 0.1]]),
numpy.array([[0., 1.], [1., 0.]])) == 0
assert error.CrossEntropyError()(numpy.array([[0.1, 1. / numpy.e],
[1. / numpy.e, 0.1]]),
numpy.array([[0., 1.], [1., 0.]])) == 1
assert error.CrossEntropyError()(numpy.array([[0.1, 1.],
[1. / numpy.e, 0.1]]),
numpy.array([[0., 1.], [1., 0.]])) == 0.5
def test_cross_entropy_derivative_equals():
"""Should not raise error or return nan, when both inputs match.
Because the derivative includes a division, this could occur.
"""
assert (list(error.CrossEntropyError().derivative(
numpy.array([0., 1.]), numpy.array([0., 1.]))[1]) == [0., -0.5])
def test_cross_entropy_zero_in_vec_a():
"""Should not raise error when zeros are in vec_a.
Because CE takes log of first vector, it can have issues with vectors containing 0s.
"""
error_func = error.CrossEntropyError()
assert error_func(numpy.array([0., 0., 1.]), numpy.array([0., 0., 1.])) == 0
assert error_func(numpy.array([1., 0., 1.]), numpy.array([0., 0., 1.])) == 0
def test_Model_stochastic_train():
"""Train with stochastic gradient descent."""
from learning import transfer, error, validation, MLP
dataset = datasets.get_iris()
model = MLP((len(dataset[0][0]), 2, len(dataset[1][0])),
transfers=transfer.SoftmaxTransfer(),
error_func=error.CrossEntropyError())
# Model should be able to converge with mini-batches
model.stochastic_train(
*dataset,
error_break=0.02,
pattern_selection_func=lambda X, Y: base.select_sample(X, Y, size=30),
train_kwargs={
'iterations': 100,
'error_break': 0.02
})
assert validation.get_error(model, *dataset) <= 0.03
def test_cross_entropy_derivative():
check_error_gradient(error.CrossEntropyError())
def test_cross_entropy_error_on_negative():
"""CrossEntropy does not take negative values in vec_a."""
error_func = error.CrossEntropyError()
with pytest.raises(FloatingPointError):
assert error_func(numpy.array([-1., 1.]), numpy.array([0., 1.]))
def test_cross_entropy_vector():
assert error.CrossEntropyError()(numpy.array([0.1, 1.]),
numpy.array([0., 1.])) == 0
assert error.CrossEntropyError()(numpy.array([0.1, 1. / numpy.e]),
numpy.array([0., 1.])) == 1
def test_cross_entropy_derivative_error_equals_call_error_matrix():
check_derivative_error_equals_call_error(error.CrossEntropyError(),
tensor_d=2)
def test_cross_entropy_derivative_matrix():
check_error_gradient(error.CrossEntropyError(), tensor_d=2)