def test_second_model_history_is_not_default(self):
n1 = 13
n2 = 23
# these will get overridden
seq = [0]
seq1_exp = [1]
seq2_exp = [2]
for i in range(2):
arma1 = Arma([0.9], [0.1, -0.2],
default_source=sources.GaussianNoise())
arma2 = Arma([0.1, -0.2], [0.3],
default_source=sources.GaussianNoise())
if i == 0:
seq1_exp = arma1.transform(n1)
seq2_exp = arma2.transform(n2)
else:
seq = sample_switching_models(
[arma1, arma2],
np.hstack((np.zeros(n1, dtype=int), np.ones(n2,
dtype=int))),
)
np.testing.assert_allclose(seq[:n1], seq1_exp)
self.assertGreater(np.max(np.abs(seq[n1:] - seq2_exp)), 0)
class TestArmaSourceScaling(unittest.TestCase):
def setUp(self):
self.source_scaling = 1.3
self.rng = np.random.default_rng(1)
self.n = 1000
self.source_data = self.rng.normal(size=self.n)
self.a = [-1.1, -0.6, -0.1]
self.b = [0.5, 0.3]
self.arma = Arma(self.a, self.b, source_scaling=self.source_scaling)
self.arma_alt = Arma(self.a, self.b, source_scaling=1)
self.y, self.x = self.arma.transform(X=self.source_data,
return_input=True)
self.y_alt, self.x_alt = self.arma_alt.transform(X=self.source_data,
return_input=True)
def test_output_scaled_by_appropriate_factor(self):
np.testing.assert_allclose(self.y, self.source_scaling * self.y_alt)
def test_input_not_scaled(self):
np.testing.assert_allclose(self.x_alt, self.x)
def test_default_scaling_is_one(self):
arma_def = Arma(self.a, self.b)
y_def = arma_def.transform(X=self.source_data)
np.testing.assert_allclose(y_def, self.y_alt)
class TestArmaMonitor(unittest.TestCase):
def setUp(self):
self.rng = np.random.default_rng(1)
self.n = 253
self.source_data = self.rng.normal(size=self.n)
self.a = [-1.1, -0.6, -0.1]
self.b = [0.5, 0.3]
self.arma = Arma(self.a, self.b)
def test_monitor_output_matches_transform_retval(self):
monitor = AttributeMonitor(["output_"])
y_out = self.arma.transform(X=self.source_data, monitor=monitor)
self.assertTrue(hasattr(monitor.history_, "output_"))
np.testing.assert_allclose(monitor.history_.output_, y_out)
def test_monitor_input_matches_actual_input(self):
monitor = AttributeMonitor(["input_"])
self.arma.transform(X=self.source_data, monitor=monitor)
self.assertTrue(hasattr(monitor.history_, "input_"))
np.testing.assert_allclose(monitor.history_.input_, self.source_data)
def test_monitor_is_initialized_on_zero_n_samples(self):
monitor = AttributeMonitor(["input_", "output_"])
self.arma.transform(0, monitor=monitor)
self.assertTrue(hasattr(monitor.history_, "input_"))
self.assertTrue(hasattr(monitor.history_, "output_"))
def test_monitor_is_initialized_on_empty_U(self):
monitor = AttributeMonitor(["input_", "output_"])
self.arma.transform(X=[], monitor=monitor)
self.assertTrue(hasattr(monitor.history_, "input_"))
self.assertTrue(hasattr(monitor.history_, "output_"))
def test_chunk_hint_does_not_affect_monitoring(self):
names = ["input_", "output_"]
monitor1 = AttributeMonitor(names)
self.arma.transform(X=self.source_data,
monitor=monitor1,
chunk_hint=11)
monitor2 = AttributeMonitor(names)
arma2 = Arma(self.a, self.b)
arma2.transform(X=self.source_data, monitor=monitor2, chunk_hint=23)
np.testing.assert_allclose(monitor1.history_.input_,
monitor2.history_.input_)
np.testing.assert_allclose(monitor1.history_.output_,
monitor2.history_.output_)
def test_chunk_hint_does_not_affect_monitoring(self):
names = ["input_", "output_"]
monitor1 = AttributeMonitor(names)
self.arma.transform(X=self.source_data,
monitor=monitor1,
chunk_hint=11)
monitor2 = AttributeMonitor(names)
arma2 = Arma(self.a, self.b)
arma2.transform(X=self.source_data, monitor=monitor2, chunk_hint=23)
np.testing.assert_allclose(monitor1.history_.input_,
monitor2.history_.input_)
np.testing.assert_allclose(monitor1.history_.output_,
monitor2.history_.output_)
def test_trivial_arma_with_bias_no_noise(self):
bias = 0.75
arma = Arma([], [], bias=bias)
y = arma.transform(X=np.zeros(15))
np.testing.assert_allclose(y, bias)
def test_inverse_recovers_input_with_nonzero_bias(self):
arma = Arma([1.3, -0.8, 0.2, -0.1], [-0.3, 0.5], bias=0.6)
inv_arma = arma.inverse()
y = arma.transform(X=self.u)
u_again = inv_arma.transform(X=y)
np.testing.assert_allclose(self.u, u_again)
def test_inverse_recovers_input_with_zero_bias(self):
arma = Arma([-1.1, -0.6, -0.1], [0.5, 0.3])
inv_arma = arma.inverse()
y = arma.transform(X=self.u)
u_again = inv_arma.transform(X=y)
np.testing.assert_allclose(self.u, u_again)
def test_ar1_zero_input(self):
alpha = 0.78
y0 = 1.0
ar = Arma([alpha], [], initial_conditions=([y0], []))
n = 10
y = ar.transform(n, X=lambda size: np.zeros(size))
y_exp = y0 * alpha**np.arange(1, n + 1)
np.testing.assert_allclose(y, y_exp)
def test_single_model(self):
n = 20
a = [0.9]
b = [0.1, -0.2]
arma1 = Arma(a, b, default_source=sources.GaussianNoise())
seq_exp = arma1.transform(n)
arma2 = Arma(a, b, default_source=sources.GaussianNoise())
seq = sample_switching_models([arma2], np.zeros(n, dtype=int))
np.testing.assert_allclose(seq, seq_exp)
class TestArmaTransformZeroSamples(unittest.TestCase):
def setUp(self):
# this also implicitly tests that this works without default_source
self.arma = Arma([1, 0.5], [-0.3])
def test_zero_samples_returns_empty_output(self):
y = self.arma.transform(0)
self.assertEqual(len(y), 0)
def test_zero_samples_returns_empty_output_and_input_when_return_input_true(
self):
y, x = self.arma.transform(0, return_input=True)
self.assertEqual(len(y), 0)
self.assertEqual(len(x), 0)
def test_empty_x_returns_empty(self):
y, x = self.arma.transform(X=[], return_input=True)
self.assertEqual(len(y), 0)
self.assertEqual(len(x), 0)
def test_asymptotic_value_constant_noise(self):
bias = 0.32
arma = Arma([-1.1, -0.6, -0.1], [0.5, 0.3], bias=bias)
# the asymptotic value y0 should be given by
# y0 * (1 - sum(a)) = bias + x0 * (1 + sum(b))
# so: y0 = (bias + x0 * (1 + sum(b))) / (1 - sum(a))
x0 = -0.5
# give it enough time to converge
n = 1000
y = arma.transform(X=x0 * np.ones(n))
y0 = (bias + x0 * (1 + np.sum(arma.b))) / (1 - np.sum(arma.a))
self.assertAlmostEqual(y[-1], y0)
def test_ma_is_convolution(self):
rng = default_rng(1)
q = 3
b = rng.normal(size=q)
ma = Arma([], b)
n = 52
x = rng.normal(size=n)
y = ma.transform(X=x)
x_padded = np.hstack((np.zeros(q), x))
b_ext = np.hstack(([1], b))
y_exp = np.convolve(x_padded, b_ext, mode="valid")
np.testing.assert_allclose(y, y_exp)
def test_initial_conditions_parameter_is_obeyed(self):
a = [0.8]
b = [0.1, 0.2]
n = 32
ic = ([-0.5], [-0.5, 0.3])
arma = Arma(a,
b,
default_source=sources.GaussianNoise(),
initial_conditions=ic)
seq_exp = arma.transform(n)
arma = Arma(a, b, default_source=sources.GaussianNoise())
seq = sample_switching_models([arma],
np.zeros(n, dtype=int),
initial_conditions=ic)
np.testing.assert_allclose(seq, seq_exp)
def test_default_scaling_is_one(self):
arma_def = Arma(self.a, self.b)
y_def = arma_def.transform(X=self.source_data)
np.testing.assert_allclose(y_def, self.y_alt)