def _states(self, V, W):
ms = [self._m0]
Cs = [self._C0]
structure = self._structure
structure._W = W # pylint: disable=protected-access
kalman = KalmanFilter(structure=structure, V=V)
for t in range(1, self._nobs):
m, C = kalman.filter(self._data[t], ms[t - 1], Cs[t - 1])
ms.append(m)
Cs.append(C)
return ms, Cs
def _states(self, V, W):
ms = [self._m0]
Cs = [self._C0]
structure = self._structure
structure._W = W
kf = KalmanFilter(structure=structure, V=V)
for t in range(1, self._nobs):
m, C = kf.filter(self._data[t], ms[t - 1], Cs[t - 1])
ms.append(m)
Cs.append(C)
return ms, Cs
def univariate_ll_bs_elements_test(self):
"""Test how many states LL BS returns"""
# Run Kalman filter
ms = [np.array([0, 0])]
Cs = [np.identity(2) * 1000]
kf = KalmanFilter(structure=FFBSTests._ll["structure"], V=1.4)
for t in range(1, FFBSTests._ll["nobs"]):
m, C = kf.filter(FFBSTests._ll["obs"][t], ms[t - 1], Cs[t - 1])
ms.append(m)
Cs.append(C)
states = FFBS.backward_sampling(ms, Cs, np.identity(2))
assert_equals(len(states), FFBSTests._ll["nobs"] - 1)
def univariate_lc_bs_dimension_test(self):
"""Test states LC BS dimension"""
# Run Kalman filter
ms = [np.array([0])]
Cs = [np.identity(1) * 1000]
kf = KalmanFilter(structure=FFBSTests._lc["structure"], V=1.4)
for t in range(1, FFBSTests._lc["nobs"]):
m, C = kf.filter(FFBSTests._lc["obs"][t], ms[t - 1], Cs[t - 1])
ms.append(m)
Cs.append(C)
states = FFBS.backward_sampling(ms, Cs, np.identity(1))
assert_equals(states[50].shape, (1, ))
def univariate_lc_filter_dimension_test(self):
"""Test if the KF output has LC dimensions."""
lc = UnivariateStructure.locally_constant(1.0)
m0 = np.array([0])
C0 = np.diag([1e3])
m, C = KalmanFilter(structure=lc, V=2.5).filter(y=0.0, m=m0, C=C0)
assert_equals(m.shape, (1, ), "First moment dimensions are wrong")
assert_equals(C.shape, (1, 1), "Second moment dimensions are wrong")
def univariate_ll_filter_dimension_test(self):
"""Test if the KF output has LL dimensions."""
ll = UnivariateStructure.locally_linear(
W=np.matrix([[0.1, 0], [0, 0.1]]))
print(ll.F.shape)
m0 = np.array([0, 0])
C0 = np.diag([1e3, 1e3])
m, C = KalmanFilter(structure=ll, V=2.5).filter(y=0.0, m=m0, C=C0)
assert_equals(m.shape, (2, ), "First moment dimensions are wrong")
assert_equals(C.shape, (2, 2), "Second moment dimensions are wrong")