def estimate_u(self, y, y_dot, gy=None):
if gy is None:
gy = generalized_gradient(y)
A, B = y.shape
n = A * B
K = self.H.shape[0]
H = self.H
# H = (K x 2 x H x W )
# gy = (2 x H x W )
Hgy = np.tensordot(H * gy, [1, 1], axes=(1, 0))
# This was the slow system
# a = np.zeros((n, K))
# b = np.zeros(n)
# s = 0
# for i, j in itertools.product(range(y.shape[0]),
# range(y.shape[1])):
# a[s, :] = Hgy[:, i, j]
# b[s] = y_dot[i, j]
# s += 1
# assert s == n
a = Hgy.reshape((K, n)).T
b = y_dot.reshape(n)
u_est, _, _, _ = np.linalg.lstsq(a, b)
return u_est
def estimate_y_dot(self, y, u, gy=None):
if gy is None:
gy = generalized_gradient(y)
H = self.H
uH = np.tensordot(u, H, axes=(0, 0))
y_dot = (uH * gy).sum(axis=0)
return y_dot
def update(self, y, y_dot, u, w):
self.once = True
M = y.shape[0]
check_all_finite(y)
check_all_finite(y_dot)
check_all_finite(u)
# TODO: check shape is conserved
self.is1D = y.ndim == 1
self.is2D = y.ndim == 2
gy = generalized_gradient(y)
y_dot_w = w
u_w = np.ones(u.shape)
gy_w = w.reshape((1, M))
assert gy.shape == gy_w.shape
Qi = outer(u, u)
Qi_w = outer(u_w, u_w)
self.Q.update(Qi, Qi_w)
Pi = outer_first_dim(gy)
Pi_w = outer_first_dim(gy_w)
self.P.update(Pi, Pi_w)
self.R_needs_update = True
Gi = outer(u, gy * y_dot)
Gi_w = outer(u_w, gy_w * y_dot_w)
self.G.update(Gi, Gi_w)
self.H_needs_update = True
Bk = outer(u, y_dot)
Bk_w = outer(u_w, y_dot_w)
self.B.update(Bk, Bk_w)
self.C_needs_update = True
self.last_y = y
self.last_gy = gy
self.last_y_dot = y_dot
self.last_u = u
self.last_w = w
def get_importance(self, y, y_dot):
self.once = True
gy = generalized_gradient(y) # gy='array[1xN]',
y_valid = np.logical_and(y > self.min_y, y < self.max_y)
gy0 = gy[0, :]
gy_valid = np.abs(gy0) < self.max_gy
y_dot_valid = np.abs(y_dot) < self.max_y_dot
w = y_valid * 1.0 * gy_valid * y_dot_valid
self.w_stats.update(w)
self.last_w = w
self.last_y = y
self.last_y_valid = y_valid
self.last_y_dot = y_dot
self.last_y_dot_valid = y_dot_valid
self.last_gy = gy
self.last_gy_valid = gy_valid
return w
