def test_reproducibility(self):
ortho = init.Orthogonal(seed=0)
weight1 = ortho.sample((10, 20), return_array=True)
weight2 = ortho.sample((10, 20), return_array=True)
np.testing.assert_array_almost_equal(weight1, weight2)
def test_orthogonal_matrix_initializer(self):
# Note: Matrix can't be orthogonal for row and column space
# at the same time in case if matrix rectangular
ortho = init.Orthogonal(scale=1)
# Matrix that have more rows than columns
weight = self.eval(ortho.sample((30, 10)))
np.testing.assert_array_almost_equal(np.eye(10),
weight.T.dot(weight),
decimal=5)
ortho = init.Orthogonal(scale=1)
# Matrix that have more columns than rows
weight = self.eval(ortho.sample((10, 30)))
np.testing.assert_array_almost_equal(np.eye(10),
weight.dot(weight.T),
decimal=5)
def test_bfgs(self):
x_train, x_test, y_train, y_test = simple_classification()
qnnet = algorithms.QuasiNewton(
connection=[
layers.Input(10),
layers.Dropout(0.1),
layers.Sigmoid(30, weight=init.Orthogonal()),
layers.Sigmoid(1, weight=init.Orthogonal()),
],
shuffle_data=True,
show_epoch='20 times',
verbose=False,
)
qnnet.train(x_train, y_train, x_test, y_test, epochs=20)
result = qnnet.predict(x_test).round().astype(int)
roc_curve_score = metrics.roc_auc_score(result, y_test)
self.assertAlmostEqual(0.92, roc_curve_score, places=2)
def test_quasi_newton_psb(self):
x_train, x_test, y_train, y_test = simple_classification()
qnnet = algorithms.QuasiNewton(
connection=[
layers.Input(10),
layers.Sigmoid(30, weight=init.Orthogonal()),
layers.Sigmoid(1, weight=init.Orthogonal()),
],
shuffle_data=True,
verbose=False,
update_function='psb',
h0_scale=2,
)
qnnet.train(x_train, y_train, x_test, y_test, epochs=3)
result = qnnet.predict(x_test).round()
roc_curve_score = metrics.roc_auc_score(result, y_test)
self.assertAlmostEqual(0.92, roc_curve_score, places=2)
def test_quasi_newton_bfgs(self):
x_train, x_test, y_train, y_test = simple_classification()
qnnet = algorithms.QuasiNewton(
network=[
layers.Input(10),
layers.Sigmoid(30, weight=init.Orthogonal()),
layers.Sigmoid(1, weight=init.Orthogonal()),
],
shuffle_data=True,
show_epoch=10,
update_function='bfgs',
)
qnnet.train(x_train, y_train, x_test, y_test, epochs=50)
result = qnnet.predict(x_test).round().astype(int)
roc_curve_score = metrics.roc_auc_score(result, y_test)
self.assertAlmostEqual(0.92, roc_curve_score, places=2)
def test_orthogonal_1d_shape(self):
ortho = init.Orthogonal(scale=1)
sampled_data = self.eval(ortho.sample(shape=(1, )))
self.assertEqual((1, ), sampled_data.shape)
def test_orthogonal_init_repr(self):
ortho_initializer = init.Orthogonal(scale=1)
self.assertEqual("Orthogonal(scale=1)", str(ortho_initializer))
def test_orthogonal_matrix_initializer_errors(self):
with self.assertRaises(ValueError):
ortho = init.Orthogonal()
# More than 2 dimensions
ortho.sample((5, 5, 5))