def sample_signal_details(self, rhs=None,
norm_order=None, max_iterations=10000,
eps=None, raise_error=True,
find_mean=False):
P, L = self.P_L
if eps is None:
eps = self.eps
if rhs is None:
rhs = self.sample_rhs(find_mean=find_mean)
self.logger.info('Starting CG iterations')
t0 = get_times()
cg_logger = make_sublogger(self.logger, 'cg')
x0 = harmonic_sphere_map(0, self.lmin, self.lmax, is_complex=False)
x, info = CG(self.cg_mul_lhs, rhs, x0=x0, precond=self.preconditioner,
norm_order=norm_order, maxit=max_iterations, eps=eps,
raise_error=raise_error, logger=cg_logger)
self.logger.info('CG done, %d its, %.2e res. (%s)',
info['iterations'],
info['residuals'][-1],
timestats(t0))
# Got x; need to scale back -- x = L^-1 sdraw
sdraw = harmonic_sphere_map(P.H * (L * (P * x)), self.lmin, self.lmax, is_complex=False)
return sdraw, info, x
def __call__(self, B):
x0 = harmonic_sphere_map(0, self.lmin, self.lmax, is_complex=False)
x, info = CG(self.cg_mul_lhs, B, x0=x0, precond=self.child,
norm_order=None, maxit=self.max_iterations,
eps=self.eps,
raise_error=True,
logger=self.cg_logger)
return x
def __call__(self, B):
X_list = []
for child, start, stop in self.child_start_stop:
sub_X = child(B[start:stop])
X_list.append(sub_X)
X = harmonic_sphere_map(np.hstack(X_list), self.lmin, self.lmax,
is_complex=False)
return X
# SPeven = butterfly(P[:, ::2])
# SPodd = butterfly(P[:, 1::2])
# print 'Compression', SPeven.size() / DenseMatrix(P[:, ::2]).size()
# print 'Compression', SPodd.size() / DenseMatrix(P[:, 1::2]).size()
if 0:
x = np.cos(get_ring_thetas(Nside))
x[np.abs(x) < 1e-10] = 0
xneg = x[x < 0]
xpos = x[x > 0]
assert np.allclose(-xneg[::-1], xpos)
X = InnerSumPerM(m, x[x >= 0], lmax)
P_even_arr = X.P_even_arr.copy('F')
P_even = X.P_even
a_l_even = a_l[::2].copy()
if 0:
map = alm2map(m, a_l, Nside)
from cmb.maps import pixel_sphere_map, harmonic_sphere_map
pixel_sphere_map(map).plot(title='fast')
alm_fid = harmonic_sphere_map(0, lmin=0, lmax=lmax, is_complex=False)
assert m != 0
for l in range(m, lmax + 1):
alm_fid[l**2 + l + m] = np.sqrt(2) * a_l[l - m] # real is repacked
print 'Diff', np.linalg.norm(map - alm_fid.to_pixel(Nside))
alm_fid.to_pixel(Nside).plot(title='fiducial')
#(map - alm_fid.to_pixel(Nside)).plot(title='diff')
def __call__(self, B):
return harmonic_sphere_map(
self.A.solve_right(B, algorithm='cholesky', check=False),
self.lmin, self.lmax, is_complex=False)
# as_matrix(np.log(np.abs(Peven))).plot()
# SPeven = butterfly(P[:, ::2])
# SPodd = butterfly(P[:, 1::2])
# print 'Compression', SPeven.size() / DenseMatrix(P[:, ::2]).size()
# print 'Compression', SPodd.size() / DenseMatrix(P[:, 1::2]).size()
if 0:
x = np.cos(get_ring_thetas(Nside))
x[np.abs(x) < 1e-10] = 0
xneg = x[x < 0]
xpos = x[x > 0]
assert np.allclose(-xneg[::-1], xpos)
X = InnerSumPerM(m, x[x >= 0], lmax)
P_even_arr = X.P_even_arr.copy('F')
P_even = X.P_even
a_l_even = a_l[::2].copy()
if 0:
map = alm2map(m, a_l, Nside)
from cmb.maps import pixel_sphere_map, harmonic_sphere_map
pixel_sphere_map(map).plot(title='fast')
alm_fid = harmonic_sphere_map(0, lmin=0, lmax=lmax, is_complex=False)
assert m != 0
for l in range(m, lmax + 1):
alm_fid[l**2 + l + m] = np.sqrt(2) * a_l[l - m] # real is repacked
print 'Diff', np.linalg.norm(map - alm_fid.to_pixel(Nside))
alm_fid.to_pixel(Nside).plot(title='fiducial')
#(map - alm_fid.to_pixel(Nside)).plot(title='diff')
