def daniels_variant(measurements, true):
iterations = 10
_, xk, P, tau, rho, u, U, H, F, Q, N = common.init_all()
nd = node_factory(xk, P, u, U, F, Q, H, rho, tau,
observe_factory(measurements.T), iterations)
nd()
preds = np.array(nd.logger['x']).squeeze()
return preds, nd.post_rmse(true.T, start_element=RMSE_START)
def make_simple_nodes(n=5, iterations=10, traj=None, noise_modifier=None):
if not noise_modifier:
noise_modifier = lambda _: 10
nodes = []
for i in range(n):
traj2, xk, P, tau, rho, u, U, H, F, Q, N = common.init_all(traj)
np.random.seed(i)
traj2.Y = traj2.Y + (np.random.normal(size=traj2.Y.shape) *
noise_modifier(traj2.Y.shape))
nd = node_factory(xk, P, u, U, F, Q, H, rho, tau,
observe_factory(traj2.Y.T.copy()), iterations)
nodes.append(nd)
return nodes
def test_single_node_rmse():
traj, xk, P, tau, rho, u, U, H, F, Q, N = init_all()
f = observe_factory(traj.Y.T)
m1 = node_factory(xk, P, u, U, F, Q, H, rho, tau, f)
m1()
x_log = np.array(m1.logger['x']).squeeze().T
out_rmse_kf = np.sqrt(((x_log[:2] - traj.X[:2])**2).mean())
out_rmse_nokf = np.sqrt(((traj.Y[:2] - traj.X[:2])**2).mean())
exp_rmse_kf = 0.4621395030767436
exp_rmse_nokf = 0.7668096325216082
np.testing.assert_almost_equal(out_rmse_kf, exp_rmse_kf)
np.testing.assert_almost_equal(out_rmse_nokf, exp_rmse_nokf)
def daniels_variant(measurements,
true,
cov,
rho=0.95,
tau=5,
u=5,
debug=False):
iterations = 10
_, xk, P, _, _, _, _, H, F, Q, N = init_all()
U = cov * (u - 3)
nd = node_factory(xk, P, u, U, F, Q, H, rho, tau,
observe_factory(measurements.T), iterations)
nd()
preds = np.array(nd.logger['x']).squeeze()
m, s = nd.rmse_stats(true.T)
if debug:
return preds, m, s, nd
return preds, m, s
def test_single_node():
traj_ = init_trajectory(5)
traj, xk, P, tau, rho, u, U, H, F, Q, N = init_all(traj_)
f = observe_factory(traj.Y.T)
m1 = node_factory(xk, P, u, U, F, Q, H, rho, tau, f)
m1()
exp_x = np.array([[-2.76740e-01, 1.09767e+00, 9.62620e-01, 1.09898e+00],
[-1.49130e-01, 2.92570e-01, 1.12011e+00, -3.49807e+00],
[1.45799e+00, 9.18870e-01, 8.70549e+00, 1.40484e+00],
[8.60280e-01, 6.99180e-01, 2.80428e+00, 3.62000e-03],
[9.53670e-01, -2.63710e-01, 2.23199e+00, -3.16430e+00]])
exp_last_P = np.array([[0.53926178, 0.03091239, 1.74759438, 0.0822738],
[0.03091239, 0.54178761, 0.08098487, 1.78451349],
[1.74759438, 0.08098487, 9.21782513, 0.24729238],
[0.0822738, 1.78451349, 0.24729238, 9.44313789]])
np.testing.assert_allclose(exp_x, np.array(m1.logger['x']), atol=1e-5)
np.testing.assert_allclose(exp_last_P, m1.logger['P'][-1], atol=1e-5)
def test_two_mean():
traj, xk, P, tau, rho, u, U, H, F, Q, N = init_all()
# observe_factory is now irrelevant
m1 = node_factory(xk, P, u, U, F, Q, H, rho, tau,
observe_factory(traj.Y.T))
m2 = node_factory(xk, P, u, U, F, Q, H, rho, tau,
observe_factory(traj.Y.T))
zs = traj.Y.T
zs_dist = traj.Y.T + np.random.normal(size=traj.Y.T.shape) * 0.1
for z1, z2 in zip(zs, zs_dist):
x1, P1, hyp_P1, hypR1 = m1.single_kf(z1)
x2, P2, hyp_P2, hypR2 = m2.single_kf(z2)
x_log = ((np.array(m1.logger['x']) + np.array(m2.logger['x'])) /
2).squeeze().T
out = np.sqrt(((x_log[:2] - traj.X[:2])**2).mean()), np.sqrt(
((traj.Y[:2] - traj.X[:2])**2).mean())
expected = (0.46242291667779106, 0.7668096325216082)
np.testing.assert_allclose(out, expected, atol=1e-3)
def make_node(measurements, cov, rho=0.9, tau=5, u=5):
_, xk, P, _, _, _, _, H, F, Q, N = init_all()
U = cov * (u - 3)
nd = node_factory(xk, P, u, U, F, Q, H, rho, tau,
observe_factory(measurements.T), 10)
return nd