def calc_conditional_prior(self, x_free, geom):
"""Calculate prior distribution of radiance. This depends on the
location in the state space. Return the inverse covariance and
its square root (for non-quadratic error residual calculation)."""
x = self.full_statevector(x_free)
xa = self.fm.xa(x, geom)
Sa = self.fm.Sa(x, geom)
# If there aren't any fixed parameters, we just directly
if self.x_fixed is None or self.grid_as_starting_points:
Sa_inv, Sa_inv_sqrt = svd_inv_sqrt(
Sa,
hashtable=self.hashtable,
max_hash_size=self.max_table_size)
return xa, Sa, Sa_inv, Sa_inv_sqrt
else:
# otherwise condition on fixed variables
#TODO: could make the below calculation without the svd_inv (using full initial inversion),
# which would be way cheaper
xa_free, Sa_free = conditional_gaussian(xa, Sa, self.inds_free,
self.inds_fixed,
self.x_fixed)
Sa_free_inv, Sa_free_inv_sqrt = svd_inv_sqrt(
Sa_free,
hashtable=self.hashtable,
max_hash_size=self.max_table_size)
return xa_free, Sa_free, Sa_free_inv, Sa_free_inv_sqrt
def calc_prior(self, x, geom):
"""Calculate prior distribution of radiance. This depends on the
location in the state space. Return the inverse covariance and
its square root (for non-quadratic error residual calculation)."""
xa = self.fm.xa(x, geom)
Sa = self.fm.Sa(x, geom)
Sa_inv, Sa_inv_sqrt = svd_inv_sqrt(Sa, hashtable=self.hashtable)
return xa, Sa, Sa_inv, Sa_inv_sqrt
def calc_Seps(self, x, meas, geom):
"""Calculate (zero-mean) measurement distribution in radiance terms.
This depends on the location in the state space. This distribution is
calculated over one or more subwindows of the spectrum. Return the
inverse covariance and its square root."""
Seps = self.fm.Seps(x, meas, geom)
wn = len(self.winidx)
Seps_win = np.zeros((wn, wn))
for i in range(wn):
Seps_win[i, :] = Seps[self.winidx[i], self.winidx]
return svd_inv_sqrt(Seps_win, hashtable=self.hashtable)
def calc_conditional_prior(self, x_free, geom):
"""Calculate prior distribution of radiance. This depends on the
location in the state space. Return the inverse covariance and
its square root (for non-quadratic error residual calculation)."""
x = self.full_statevector(x_free)
xa = self.fm.xa(x, geom)
Sa = self.fm.Sa(x, geom)
# If there aren't any fixed parameters, we just directly
if self.x_fixed is None:
x[self.inds_fixed] = self.x_fixed
Sa_inv, Sa_inv_sqrt = svd_inv_sqrt(Sa, hashtable=self.hashtable)
return xa, Sa, Sa_inv, Sa_inv_sqrt
else:
# otherwise condition on fixed variables
xa_free, Sa_free = conditional_gaussian(xa, Sa, self.inds_free,
self.inds_fixed,
self.x_fixed)
Sa_free_inv, Sa_free_inv_sqrt = svd_inv_sqrt(
Sa_free, hashtable=self.hashtable)
return xa_free, Sa_free, Sa_free_inv, Sa_free_inv_sqrt
