def test_gradient_checking(self):
layer = Loss('binary_crossentropy')
X = np.random.rand(5, 1)
y = np.random.randint(0, high=2, size=(5, ))
expected = approx_fprime(X, layer.forwardprop, y)
got = layer.backprop(X, y)
np.testing.assert_array_almost_equal(got, expected, decimal=5)
def test_gradient_checking(self):
layer = Loss('mse')
X = np.random.rand(5, 1)
y = np.random.rand(5, )
expected = approx_fprime(X, layer.forwardprop, y)
got = layer.backprop(X, y)
np.testing.assert_array_almost_equal(got, expected, decimal=6)
def test_gradient_checking(self):
layer = Dropout(0.36)
X = np.random.randn(5, 4)
dout = np.random.randn(5, 4)
layer.forwardprop(X, training=True)
expected = approx_fprime(X, layer.forwardprop, True, False)
got = layer.backprop(dout)
np.testing.assert_array_almost_equal(got / dout, expected, decimal=6)
def test_gradient_checking(self):
layer = Activation('tanh')
X = np.random.randn(5, 4)
expected = approx_fprime(X, layer.activation.forwardprop)
got = layer.activation.backprop(X)
np.testing.assert_array_almost_equal(got, expected, decimal=5)