def imf_calc(self):
theta = self.const.theta_0
M = np.zeros((self.const.s, self.const.s))
for k in range(self.const.N):
print(f"WE SURVIVED {k} ITERATIONS")
self.const.a = np.dot(self.model.Psi(theta), self.const.U[k])
a_gr = self.a_gr(k)
if k == 0:
x_A_tk1 = np.reshape([np.dot(self.model.F(theta), self.model.mu_x(theta)) + self.const.a] + \
[np.dot(self.model.F_grad(theta, i), self.model.mu_x(theta)) +
np.dot(self.model.F(theta), self.model.mu_x_grad(theta, i)) +
a_gr[i]
for i in range(self.const.s)], (self.const.s + 1, self.const.m))
sigma_A_tk1 = np.zeros((self.const.s + 1, self.const.s + 1))
E_x_hat = x_A_tk1
E_x_hat_x_hat_t = sigma_A_tk1 + np.dot(x_A_tk1, T(x_A_tk1))
for i in range(self.const.s):
for j in range(self.const.s):
M[i][j] += TCD(self.model.H_grad(theta, i), self.C(0), E_x_hat_x_hat_t, T(self.C(0)),
T(self.model.H_grad(theta, j)), inv(self.kalman.B(theta))) + \
TCD(self.model.H_grad(theta, i), self.C(0), E_x_hat_x_hat_t, T(self.C(j)),
T(self.model.H(theta)), inv(self.kalman.B(theta))) + \
TCD(self.model.H_grad(theta, i), self.C(0), E_x_hat, T(self.model.A_grad(theta, j)),
inv(self.kalman.B(theta))) + \
TCD(self.model.H(theta), self.C(i), E_x_hat_x_hat_t, T(self.C(0)),
T(self.model.H_grad(theta, j)),
inv(self.kalman.B(theta))) + \
TCD(self.model.H(theta), self.C(i), E_x_hat_x_hat_t, T(self.C(j)),
T(self.model.H(theta)),
inv(self.kalman.B(theta))) + \
TCD(self.model.H(theta), self.C(i), E_x_hat, T(self.model.A_grad(theta, j)),
inv(self.kalman.B(theta))) + \
TCD(self.model.A_grad(theta, i), T(E_x_hat), T(self.C(0)),
T(self.model.H_grad(theta, i)),
inv(self.kalman.B(theta))) + \
TCD(self.model.A_grad(theta, i), T(E_x_hat), T(self.C(j)), T(self.model.H(theta)),
inv(self.kalman.B(theta))) + \
TCD(self.model.A_grad(theta, i), T(self.model.A_grad(theta, j)),
inv(self.kalman.B(theta))) + \
0.5 * TCD(self.kalman.B_tk1_grad(theta, i), inv(self.kalman.B(theta)),
self.kalman.B_tk1_grad(theta, j), inv(self.kalman.B(theta)))
if k != self.const.N - 1:
x_A_tk1 = np.reshape(
(T(np.dot(self.F_A_tk(), x_A_tk1)) + self.a_A_tk()),
(self.const.s + 1, self.const.m))
sigma_A_tk1 = combo_dot(
self.F_A_tk(), sigma_A_tk1, T(self.F_A_tk())) + combo_dot(
self.K_A_tk(), self.kalman.B(theta), T(self.K_A_tk()))
a = np.hstack(self.kalman.P_tk1_tk1(theta))[0]
b = np.hstack([
self.kalman.P_tk1_tk1_grad(theta, i)
for i in range(self.const.s)
])[0]
self.kalman.model.P_tk_tk = lambda theta: a
self.kalman.model.P_tk_tk_grad = lambda theta, i: b[i]
return M
def K(self, theta):
return np.dot(np.dot(self.P_tk1_tk(theta), T(self.model.H(theta))), inv(self.B(theta)))
def K_tk1_grad(self, theta, i):
return np.dot(np.dot(self.P_tk1_tk_grad(theta, i), T(self.model.H(theta))) +
np.dot(self.P_tk1_tk(theta), T(self.model.H_grad(theta, i))) -
np.dot(np.dot(np.dot(self.P_tk1_tk(theta), T(self.model.H(theta))), inv(self.B(theta))),
self.B_tk1_grad(theta, i)), inv(self.B(theta)))
def test_T_1():
assert np.array_equal(inv(2), 0.5)
def test_T_5():
assert np.array_equal(inv(npa), npa_inv)
def test_T_4():
assert np.array_equal(inv(a), a_inv)
def test_T_3():
assert np.array_equal(inv([1, 2]), [1, 0.5])
def test_T_2():
assert np.array_equal(inv([2]), [0.5])