def compute_coupling_coefficients(self,
fla1,
fla2,
flb1=None,
flb2=None,
lmax=None,
n_iter=3,
l_toeplitz=-1,
l_exact=-1,
dl_band=-1):
"""
Computes coupling coefficients of the Gaussian covariance \
between the power spectra of two pairs of NmtField objects \
(fla1, fla2, flb1 and flb2). Note that you can reuse this \
workspace for the covariance of power spectra between any \
pairs of fields as long as the fields have the same masks \
as those passed to this function, and as long as the binning \
scheme used are also the same.
:param NmtField fla1,fla2: fields contributing to the first \
power spectrum whose covariance you want to compute.
:param NmtField flb1,flb2: fields contributing to the second \
power spectrum whose covariance you want to compute. If \
None, fla1,fla2 will be used.
:param n_iter: number of iterations when computing a_lms.
:param l_toeplitz: if a positive number, the Toeplitz approximation \
described in Louis et al. 2020 (arXiv:2010.14344) will be used. \
In that case, this quantity corresponds to ell_toeplitz in Fig. \
3 of that paper.
:param l_exact: if `l_toeplitz>0`, this quantity corresponds to \
ell_exact in Fig. 3 of Louis et al. 2020. Ignored if \
`l_toeplitz<=0`.
:param dl_band: if `l_toeplitz>0`, this quantity corresponds to \
Delta ell_band in Fig. 3 of Louis et al. 2020. Ignored if \
`l_toeplitz<=0`.
"""
if flb1 is None:
flb1 = fla1
if flb2 is None:
flb2 = fla2
if self.wsp is not None:
lib.covar_workspace_free(self.wsp)
self.wsp = None
ns = fla1.fl.cs.n_eq
if (fla2.fl.cs.n_eq != ns) or (flb1.fl.cs.n_eq != ns) or \
(flb2.fl.cs.n_eq != ns):
raise ValueError("Everything should have the same resolution!")
if lmax is None:
lmax = lib.get_lmax_from_cs_py(fla1.fl.cs)
_toeplitz_sanity(l_toeplitz, l_exact, dl_band, lmax, fla1, flb1)
self.wsp = lib.covar_workspace_init_py(fla1.fl, fla2.fl, flb1.fl,
flb2.fl, lmax, n_iter,
l_toeplitz, l_exact, dl_band)
def compute_coupling_matrix(self,
fl1,
fl2,
bins,
is_teb=False,
n_iter=3,
lmax_mask=-1,
l_toeplitz=-1,
l_exact=-1,
dl_band=-1):
"""
Computes coupling matrix associated with the cross-power spectrum \
of two NmtFields and an NmtBin binning scheme. Note that the mode \
coupling matrix will only contain ells up to the maximum multipole \
included in the NmtBin bandpowers.
:param NmtField fl1,fl2: fields to correlate
:param NmtBin bin: binning scheme
:param boolean is_teb: if true, all mode-coupling matrices \
(0-0,0-2,2-2) will be computed at the same time. In this case, \
fl1 must be a spin-0 field and fl1 must be spin-2.
:param n_iter: number of iterations when computing a_lms.
:param lmax_mask: maximum multipole for masks. If smaller than the \
maximum multipoles of the fields, it will be set to that.
:param l_toeplitz: if a positive number, the Toeplitz approximation \
described in Louis et al. 2020 (arXiv:2010.14344) will be used. \
In that case, this quantity corresponds to ell_toeplitz in Fig. \
3 of that paper.
:param l_exact: if `l_toeplitz>0`, this quantity corresponds to \
ell_exact in Fig. 3 of Louis et al. 2020. Ignored if \
`l_toeplitz<=0`.
:param dl_band: if `l_toeplitz>0`, this quantity corresponds to \
Delta ell_band in Fig. 3 of Louis et al. 2020. Ignored if \
`l_toeplitz<=0`.
"""
if self.wsp is not None:
lib.workspace_free(self.wsp)
self.wsp = None
_toeplitz_sanity(l_toeplitz, l_exact, dl_band, bins.bin.ell_max, fl1,
fl2)
self.wsp = lib.comp_coupling_matrix(fl1.fl, fl2.fl, bins.bin,
int(is_teb), int(n_iter),
lmax_mask, l_toeplitz, l_exact,
dl_band)
def compute_full_master(f1,
f2,
b,
cl_noise=None,
cl_guess=None,
workspace=None,
n_iter=3,
lmax_mask=-1,
l_toeplitz=-1,
l_exact=-1,
dl_band=-1):
"""
Computes the full MASTER estimate of the power spectrum of two \
fields (f1 and f2). This is equivalent to successively calling:
- :func:`pymaster.NmtWorkspace.compute_coupling_matrix`
- :func:`pymaster.deprojection_bias`
- :func:`pymaster.compute_coupled_cell`
- :func:`pymaster.NmtWorkspace.decouple_cell`
:param NmtField f1,f2: fields to correlate
:param NmtBin b: binning scheme defining output bandpower
:param cl_noise: noise bias (i.e. angular power spectrum of \
masked noise realizations) (optional).
:param cl_guess: set of power spectra corresponding to a \
best-guess of the true power spectra of f1 and f2. Needed \
only to compute the contaminant cleaning bias (optional).
:param NmtWorkspace workspace: object containing the mode-coupling \
matrix associated with an incomplete sky coverage. If \
provided, the function will skip the computation of the \
mode-coupling matrix and use the information encoded in this \
object.
:param n_iter: number of iterations when computing a_lms.
:param lmax_mask: maximum multipole for masks. If smaller than the \
maximum multipoles of the fields, it will be set to that.
:param l_toeplitz: if a positive number, the Toeplitz approximation \
described in Louis et al. 2020 (arXiv:2010.14344) will be used. \
In that case, this quantity corresponds to ell_toeplitz in Fig. \
3 of that paper.
:param l_exact: if `l_toeplitz>0`, this quantity corresponds to \
ell_exact in Fig. 3 of Louis et al. 2020. Ignored if \
`l_toeplitz<=0`.
:param dl_band: if `l_toeplitz>0`, this quantity corresponds to \
Delta ell_band in Fig. 3 of Louis et al. 2020. Ignored if \
`l_toeplitz<=0`.
:return: set of decoupled bandpowers
"""
if f1.fl.cs.n_eq != f2.fl.cs.n_eq:
raise ValueError("Fields must have same resolution")
if cl_noise is not None:
if len(cl_noise) != f1.fl.nmaps * f2.fl.nmaps:
raise ValueError("Wrong length for noise power spectrum")
cln = cl_noise.copy()
else:
cln = np.zeros([f1.fl.nmaps * f2.fl.nmaps, 3 * f1.fl.cs.n_eq])
if cl_guess is not None:
if len(cl_guess) != f1.fl.nmaps * f2.fl.nmaps:
raise ValueError("Wrong length for guess power spectrum")
clg = cl_guess.copy()
else:
clg = np.zeros([f1.fl.nmaps * f2.fl.nmaps, 3 * f1.fl.cs.n_eq])
_toeplitz_sanity(l_toeplitz, l_exact, dl_band, b.bin.ell_max, f1, f2)
if workspace is None:
cl1d = lib.comp_pspec(f1.fl, f2.fl, b.bin, None, cln, clg,
len(cln) * b.bin.n_bands, n_iter, lmax_mask,
l_toeplitz, l_exact, dl_band)
else:
cl1d = lib.comp_pspec(f1.fl, f2.fl, b.bin, workspace.wsp, cln, clg,
len(cln) * b.bin.n_bands, n_iter, lmax_mask,
l_toeplitz, l_exact, dl_band)
clout = np.reshape(cl1d, [len(cln), b.bin.n_bands])
return clout