def test_cov_mmarg():
sm = 0.1
data = get_config(dtype0='galaxy_shear')
data['tracers']['Dummy__0']['sigma_m'] = sm
# Homemade marginalized covariance
# First, get decoupled power spectra
# Theory power spectra
clf = ClFid(data, 'Dummy__0', 'Dummy__0')
_, cl = clf.get_ell_cl()
shutil.rmtree(tmpdir1)
# Binning
clc = Cl(data, 'Dummy__0', 'Dummy__0')
wp = clc.get_bandpower_windows()
shutil.rmtree(tmpdir1)
ncl, nbpw, _, nl = wp.shape
wp = wp.reshape((ncl*nbpw, ncl*nl))
cl = cl.reshape(ncl*nl)
cl = np.dot(wp, cl)
# Marginalized covariance term
covmargb = 4*sm**2*cl[:, None]*cl[None, :]
# Do with xCell
covc = Cov(data, 'Dummy__0', 'Dummy__0', 'Dummy__0', 'Dummy__0')
covmarg = covc.get_covariance_m_marg()
shutil.rmtree(tmpdir1)
assert np.amax(np.fabs(covmarg-covmargb))/np.mean(covmarg) < 1E-5
def test_get_covariance(cldata):
# Get cl from randomnly generated map ("data")
config = get_config(fsky=1)
config['cov']['cls_from_data'] = cldata
cl_class = Cl(config, 'Dummy__0', 'Dummy__0')
ell, cl_data = cl_class.get_ell_cl()
# Get cl from mapper (the true one)
m1, m2 = cl_class.get_mappers()
w = cl_class.get_workspace()
cl_m1 = m1.get_cl()
cl_m1_cp = w.couple_cell([cl_m1])
cl_m1 = w.decouple_cell(cl_m1_cp)
# # With no mask, there should not be any coupling
# rdev = cl_m1_cp / cl_m1 - 1
# assert np.max(np.abs(rdev) < 1e-5)
# Compute covariance
cov_class = Cov(config, 'Dummy__0', 'Dummy__0', 'Dummy__0', 'Dummy__0')
cov = cov_class.get_covariance()
diag = (2 * cl_m1[0]**2) / (2 * ell + 1) / 4
cov_m = np.diag(diag)
icov = np.linalg.inv(cov)
icov_m = np.linalg.inv(cov_m)
dCl = (cl_data - cl_m1)[0]
chi2 = dCl.dot(icov).dot(dCl)
chi2_m = dCl.dot(icov_m).dot(dCl)
shutil.rmtree(tmpdir1)
assert np.fabs(chi2/chi2_m-1) < 0.03
def test_custom_auto():
# No custom auto
data = get_config()
clc1 = Cl(data, 'Dummy__0', 'Dummy__0')
l1, cl1 = clc1.get_ell_cl_cp()
shutil.rmtree(tmpdir1)
# With custom auto
data = get_config()
data['tracers']['Dummy__0']['custom_auto'] = True
data['tracers']['Dummy__0']['custom_offset'] = np.pi*1E-5
clc2 = Cl(data, 'Dummy__0', 'Dummy__0')
l2, cl2 = clc2.get_ell_cl_cp()
shutil.rmtree(tmpdir1)
assert np.allclose(cl1, cl2-np.pi*1E-5, rtol=1E-4, atol=0)
# Covariance custom cross
data = get_config()
data['tracers']['Dummy__0']['custom_auto'] = True
data['tracers']['Dummy__0']['custom_offset'] = np.pi*1E-5
clc3 = Cl(data, 'Dummy__0', 'Dummy__0')
l2, cl3 = clc3.get_ell_cl_cp_cov()
shutil.rmtree(tmpdir1)
assert np.allclose(cl1, cl3, rtol=1E-4, atol=0)
def test_cl_correction():
data = get_config()
cl_class = Cl(data, 'Dummy__2', 'Dummy__2')
cl_file = cl_class.get_cl_file()
correct = cl_file['correction']
dummy = MapperDummy(data['tracers']['Dummy__2'])
w_a = dummy.get_mask()
w_b = dummy.get_mask()
n_a = dummy.mask_power
n_b = dummy.mask_power
correct_b = np.mean(w_a*w_b)/np.mean(w_a**n_a*w_b**n_b)
assert correct != 1
assert correct == correct_b
def test_file_inconsistent_errors():
clo = get_cl_class()
ell, cl = clo.get_ell_cl()
# Change bpws and try to read file
data = get_config(0.2)
data['bpw_edges'] = data['bpw_edges'][:-1]
data['recompute']['cls'] = False
data['recompute']['mcm'] = False
os.remove(os.path.join(tmpdir1, 'data.yml'))
clo2 = Cl(data, 'Dummy__0', 'Dummy__0')
with pytest.raises(ValueError):
clo2.get_ell_cl()
shutil.rmtree(tmpdir1)
def test_ignore_existing_yml():
# Test for Cls
cl_class = get_cl_class()
data = cl_class.data.data
# Now, data['cls']['Dummy-Dummy']['compute'] = 'all'. We change it to
# 'auto' and check that is read when ignore_existing_yml=True
data['cls']['Dummy-Dummy']['compute'] = 'auto'
cl_class01 = Cl(data, 'Dummy__0', 'Dummy__1', ignore_existing_yml=True)
assert cl_class01.data.data['cls']['Dummy-Dummy']['compute'] == 'auto'
cl_class01 = Cl(data, 'Dummy__0', 'Dummy__1', ignore_existing_yml=False)
assert cl_class01.data.data['cls']['Dummy-Dummy']['compute'] == 'all'
# Test for Cov
cov_class = Cov(data, 'Dummy__0', 'Dummy__1', 'Dummy__0', 'Dummy__1',
ignore_existing_yml=True)
assert cov_class.data.data['cls']['Dummy-Dummy']['compute'] == 'auto'
cov_class = Cov(data, 'Dummy__0', 'Dummy__1', 'Dummy__0', 'Dummy__1',
ignore_existing_yml=False)
assert cov_class.data.data['cls']['Dummy-Dummy']['compute'] == 'all'
shutil.rmtree(tmpdir1)
def test_ell_cl_autocov(use):
s = get_ClSack(use)
data = get_data()
for dtype in s.s.get_data_types():
for trs in s.s.get_tracer_combinations(dtype):
if dtype in ['cl_00', 'cl_0e', 'cl_ee']:
ix = 0
elif dtype in ['cl_0b', 'cl_eb']:
ix = 1
elif dtype in ['cl_be']:
ix = 2
elif dtype in ['cl_bb']:
ix = 3
else:
raise ValueError(f'data_type {dtype} must be weird!')
if use == 'fiducial':
cl_class = ClFid(data.data, *trs)
else:
cl_class = Cl(data.data, *trs)
if use == 'nl':
ell, cl = s.s.get_ell_cl(dtype, trs[0], trs[1])
ell2, cl2 = cl_class.get_ell_nl()
else:
ell, cl, cov = s.s.get_ell_cl(dtype,
trs[0],
trs[1],
return_cov=True)
ell2, cl2 = cl_class.get_ell_cl()
cov_class = Cov(data.data, *trs, *trs)
nbpw = ell.size
ncls = cl2.shape[0]
cov2 = cov_class.get_covariance()
cov2 = cov2.reshape((nbpw, ncls, nbpw, ncls))[:, ix, :, ix]
assert np.max(np.abs((cov - cov2) / np.mean(cov))) < 1e-5
if use == 'fiducial':
# Matrices to bin the fiducial Cell
ws_bpw = np.zeros((ell.size, ell2.size))
ws_bpw[np.arange(ell.size), ell.astype(int)] = 1
assert np.all(cl == ws_bpw.dot(cl2[ix]))
else:
assert np.all(cl == cl2[ix])
assert np.all(ell == ell2)
def test_get_nmtbin():
# This test checks that the bpw_edges can be read from the
# global part of the yaml file or from the context of one
# of the cross-correlations.
# 1. From global
data = get_config()
cl1 = Cl(data, 'Dummy__0', 'Dummy__0')
shutil.rmtree(tmpdir1)
# 2. From cross-correlations
data['cls']['Dummy-Dummy']['bpw_edges'] = data.pop('bpw_edges')
cl2 = Cl(data, 'Dummy__0', 'Dummy__0')
shutil.rmtree(tmpdir1)
# Check they are the same
b1 = cl1.get_NmtBin()
b2 = cl2.get_NmtBin()
assert np.all(b1.get_effective_ells() == b2.get_effective_ells())
def get_cl_class(fsky=0.2, fiducial=False):
data = get_config(fsky)
if fiducial:
return ClFid(data, 'Dummy__0', 'Dummy__0')
else:
return Cl(data, 'Dummy__0', 'Dummy__0')
def test_symmetric():
data = get_config()
# Request only 'auto' to test if read_symmetric works in the case you have
# 'auto but you need the cross for the covariance
data['cls']['Dummy-Dummy']['compute'] = 'auto'
cl_class01 = Cl(data, 'Dummy__0', 'Dummy__1')
os.remove(os.path.join(tmpdir1, 'data.yml'))
cl_class10 = Cl(data, 'Dummy__1', 'Dummy__0')
fname = os.path.join(cl_class10.outdir, 'cl_Dummy__1_Dummy__0.npz')
assert not os.path.isfile(fname)
fname = os.path.join(cl_class10.outdir, 'w__mask_dummy1_mask_dummy0.fits')
assert not os.path.isfile(fname)
assert np.all(np.array(cl_class01.get_masks()) ==
np.array(cl_class10.get_masks()[::-1]))
assert np.all(cl_class01.get_ell_cl()[1] == cl_class10.get_ell_cl()[1])
assert np.all(cl_class01.get_ell_nl()[1] == cl_class10.get_ell_nl()[1])
assert np.all(cl_class01.get_ell_nl_cp()[1] ==
cl_class10.get_ell_nl_cp()[1])
shutil.rmtree(tmpdir1)
def test_cov_ng_1h():
# From CCL directly
data = get_config(fsky=0.2)
clc = Cl(data, 'Dummy__0', 'Dummy__0')
ells = clc.b.get_effective_ells()
shutil.rmtree(tmpdir1)
cosmo = ccl.Cosmology(**data['cov']['fiducial']['cosmo'])
md = ccl.halos.MassDef200m()
mf = ccl.halos.MassFuncTinker10(cosmo, mass_def=md)
hb = ccl.halos.HaloBiasTinker10(cosmo, mass_def=md)
cm = ccl.halos.ConcentrationDuffy08(mdef=md)
hmc = ccl.halos.HMCalculator(cosmo, mf, hb, md)
pr = ccl.halos.HaloProfileHOD(cm, lMmin_0=12.1,
lM1_p=0.1, bg_0=1.2)
prof2pt = ccl.halos.Profile2ptHOD()
z, nz = np.loadtxt('xcell/tests/data/DESY1gc_dndz_bin0.txt',
usecols=(1, 3), unpack=True)
tr = ccl.NumberCountsTracer(cosmo, False, dndz=(z, nz),
bias=(z, np.ones_like(z)))
k_arr = np.geomspace(1E-4, 1E2, 256)
a_arr = 1./(1+np.linspace(0, 3, 15)[::-1])
tkk = ccl.halos.halomod_Tk3D_1h(cosmo, hmc,
prof1=pr, prof2=pr, prof12_2pt=prof2pt,
prof3=pr, prof4=pr, prof34_2pt=prof2pt,
normprof1=True, normprof2=True,
normprof3=True, normprof4=True,
a_arr=a_arr, lk_arr=np.log(k_arr))
covNG0 = ccl.angular_cl_cov_cNG(cosmo,
cltracer1=tr, cltracer2=tr,
ell=ells, tkka=tkk, fsky=1.,
cltracer3=tr, cltracer4=tr, ell2=ells)
# Gaussian only
data = get_config(fsky=0.2, inc_hm=True)
data['tracers']['Dummy__0']['hod_params'] = {'lMmin_0': 12.1,
'lM1_p': 0.1,
'bg_0': 1.2}
covcG = Cov(data, 'Dummy__0', 'Dummy__0', 'Dummy__0', 'Dummy__0')
covG = covcG.get_covariance()
shutil.rmtree(tmpdir1)
# Gaussian + non-Gaussian
data = get_config(fsky=0.2, inc_hm=True)
data['cov']['non_Gaussian'] = True
data['cov']['NG_terms'] = ['1h']
data['tracers']['Dummy__0']['hod_params'] = {'lMmin_0': 12.1,
'lM1_p': 0.1,
'bg_0': 1.2}
covc1 = Cov(data, 'Dummy__0', 'Dummy__0', 'Dummy__0', 'Dummy__0')
mapper = MapperDummy(data['tracers']['Dummy__0'])
fsky = np.mean((mapper.get_mask() > 0))
covNG1 = covc1.get_covariance_ng_halomodel(0, 0, 0, 0, fsky)
cov1 = covc1.get_covariance()
shutil.rmtree(tmpdir1)
# fsky on input
data = get_config(fsky=0.2, inc_hm=True)
data['cov']['non_Gaussian'] = True
data['cov']['NG_terms'] = ['1h']
data['cov']['fsky_NG'] = 0.1
data['tracers']['Dummy__0']['hod_params'] = {'lMmin_0': 12.1,
'lM1_p': 0.1,
'bg_0': 1.2}
covc2 = Cov(data, 'Dummy__0', 'Dummy__0', 'Dummy__0', 'Dummy__0')
covNG2 = covc2.get_covariance()-covG
shutil.rmtree(tmpdir1)
# Tests
# Compare result of NG method with G+NG-G
assert np.allclose(covNG1, cov1-covG, atol=0)
# Compare with CCL prediction
# (interpolation errors are ~1E-4)
assert np.allclose(covNG0, covNG1*fsky, atol=0, rtol=1E-3)
# fsky scaling
assert np.allclose(covNG2, covNG1*fsky/0.1, atol=0)