def get_round_obs(self, obs, gm0round):
"""
get round version of obs, with simulated galaxy image
set jacobian center to current best center, to simplify
fitting later
"""
# a copy
jacob=obs.get_jacobian()
psf_round=obs.psf.gmix.make_round()
gm_round = gm0round.convolve( psf_round )
im_nonoise=gm_round.make_image(obs.image.shape,
jacobian=jacob)
#noise=1.0/sqrt( median(obs.weight) )
noise=1.0/sqrt( obs.weight.max() )
nim = numpy.random.normal(scale=noise, size=im_nonoise.shape)
im = im_nonoise + nim
psf_obs=Observation(zeros( (1,1) ), gmix=psf_round)
# note this is different from roundify in ngmix where
# the gmix is the unconvolved one
new_obs=Observation(im,
gmix=gm_round,
weight=obs.weight.copy(),
jacobian=jacob,
psf=psf_obs)
new_obs.im_nonoise=im
return new_obs
def _get_band_observations(self, band, mindex):
"""
Get an ObsList for the coadd observations in each band
If psf fitting fails, the ObsList will be zero length
note we have already checked that we have a coadd and a single epoch
without flags
"""
meds=self.meds_list[band]
ncutout=meds['ncutout'][mindex]
image_flags=self._get_image_flags(band, mindex)
coadd_obs_list = ObsList()
obs_list = ObsList()
fake=numpy.zeros((0,0))
for icut in xrange(ncutout):
flags=0
if not self._should_use_obs(band, mindex, icut):
obs = Observation(fake)
flags = IMAGE_FLAGS
else:
iflags = image_flags[icut]
if iflags != 0:
flags = IMAGE_FLAGS
obs = Observation(fake)
else:
obs = self._get_band_observation(band, mindex, icut)
if obs is None:
flags = IMAGE_FLAGS
obs = Observation(fake)
# fill the meta data
self._fill_obs_meta_data(obs,band,mindex,icut)
# set flags
meta = {'flags':flags}
obs.update_meta_data(meta)
if icut==0:
coadd_obs_list.append(obs)
else:
obs_list.append(obs)
if self.conf['reject_outliers'] and len(obs_list) > 0:
self._reject_outliers(obs_list)
obs_list.update_meta_data({'band_num':band})
coadd_obs_list.update_meta_data({'band_num':band})
return coadd_obs_list, obs_list
def _get_band_observations(self, band, mindex):
"""
Get an ObsList for the coadd observations in each band
If psf fitting fails, the ObsList will be zero length
note we have already checked that we have a coadd and a single epoch
without flags
"""
meds = self.meds_list[band]
ncutout = meds["ncutout"][mindex]
image_flags = self._get_image_flags(band, mindex)
coadd_obs_list = ObsList()
obs_list = ObsList()
for icut in xrange(ncutout):
iflags = image_flags[icut]
if iflags != 0:
flags = IMAGE_FLAGS
obs = Observation(numpy.zeros((0, 0)))
else:
obs = self._get_band_observation(band, mindex, icut)
flags = 0
# fill the meta data
self._fill_obs_meta_data(obs, band, mindex, icut)
# set flags
meta = {"flags": flags}
obs.update_meta_data(meta)
if icut == 0:
coadd_obs_list.append(obs)
else:
obs_list.append(obs)
"""
if ncutout == 0:
for o in [coadd_obs_list,obs_list]:
flags = IMAGE_FLAGS
obs = Observation(numpy.zeros((0,0)))
meta = {'flags':flags}
obs.update_meta_data(meta)
o.append(obs)
"""
if self.conf["reject_outliers"] and len(obs_list) > 0:
self._reject_outliers(obs_list)
return coadd_obs_list, obs_list
def _trim_images(self, mbo_input, censky):
"""
cen in sky coords, relative to jacobian
center
"""
mbo = MultiBandObsList()
for obslist in mbo_input:
new_obslist=ObsList()
for obs in obslist:
j=obs.jacobian
scale=j.get_scale()
cenpix=array( j.get_cen() )
new_cen = cenpix + censky/scale
#print("cen0:",cenpix,"newcen:",new_cen)
new_im=_trim_image(obs.image, new_cen)
new_wt=_trim_image(obs.weight, new_cen)
new_j = j.copy()
new_j.set_cen(row=new_cen[0], col=new_cen[1])
newobs = Observation(
new_im,
weight=new_wt,
jacobian=new_j,
psf=obs.psf,
)
new_obslist.append( newobs )
mbo.append( new_obslist )
return mbo
def test_gaussmom_flags():
"""
test we get flags for very noisy data
"""
rng = np.random.RandomState(seed=100)
ntrial = 10
noise = 100000
scale = 0.263
dims = [32] * 2
weight = np.zeros(dims) + 1.0 / noise**2
cen = (np.array(dims) - 1) / 2
jacobian = ngmix.DiagonalJacobian(row=cen[0], col=cen[1], scale=scale)
flags = np.zeros(ntrial)
for i in range(ntrial):
im = rng.normal(scale=noise, size=dims)
obs = Observation(
image=im,
weight=weight,
jacobian=jacobian,
)
fitter = GaussMom(fwhm=1.2)
res = fitter.go(obs)
flags[i] = res['flags']
assert np.any(flags != 0)
def test_gmix_loglike_fdiff(start):
"""
test that the approx fdiff calculation agrees well enough with that
calculated with the exp() value
restrict comparisons to the valid range, which is +/- 5 sigma
"""
rtol = 4.0e-5
row, col, g1, g2, T, flux = -0.5, -0.4, 0, 0, 2.8, 5
sigma = np.sqrt(T / 2)
pars = np.array([row, col, g1, g2, T, flux])
gm = make_gmix_model(pars, 'gauss')
jacob = DiagonalJacobian(row=6.5, col=7.1, scale=0.25)
area = jacob.area
rng = np.random.RandomState(seed=10)
dims = (13, 15)
obs = Observation(
image=rng.normal(size=dims),
weight=np.exp(rng.normal(size=dims)),
jacobian=jacob,
)
fdiff = np.zeros(dims[0] * dims[1] + start, dtype=np.float64)
gm.fill_fdiff(obs, fdiff, start=start)
cen = (np.array(dims) - 1) / 2
row_range = int(cen[0] - 2 * sigma), int(cen[0] + 2 * sigma) + 1
col_range = int(cen[1] - 2 * sigma), int(cen[1] + 2 * sigma) + 1
loc = start
pnorm = 5 / 2.0 / np.pi / sigma**2
for r in range(row_range[0], row_range[1]):
for c in range(col_range[0], col_range[1]):
loc = start + r * dims[1] + c
v, u = obs.jacobian(r, c)
chi2 = ((u - col)**2 + (v - row)**2) / sigma**2
model = pnorm * np.exp(-0.5 * chi2) * area
_fdiff = (model - obs.image[r, c]) * np.sqrt(obs.weight[r, c])
print(_fdiff, fdiff[loc])
assert np.allclose(_fdiff, fdiff[loc], rtol=rtol)
def _get_band_observations(self, band, mindex):
"""
Get an ObsList for the coadd observations in each band
If psf fitting fails, the ObsList will be zero length
note we have already checked that we have a coadd and a single epoch
without flags
"""
meds = self.meds_list[band]
ncutout = meds['ncutout'][mindex]
image_flags = self._get_image_flags(band, mindex)
coadd_obs_list = ObsList()
obs_list = ObsList()
fake = numpy.zeros((0, 0))
for icut in xrange(ncutout):
flags = 0
if not self._should_use_obs(band, mindex, icut):
obs = Observation(fake)
flags = IMAGE_FLAGS
else:
iflags = image_flags[icut]
if iflags != 0:
flags = IMAGE_FLAGS
obs = Observation(fake)
else:
obs = self._get_band_observation(band, mindex, icut)
if obs is None:
flags = IMAGE_FLAGS
obs = Observation(fake)
# fill the meta data
self._fill_obs_meta_data(obs, band, mindex, icut)
# set flags
meta = {'flags': flags}
obs.update_meta_data(meta)
if icut == 0:
coadd_obs_list.append(obs)
else:
obs_list.append(obs)
if self.conf['reject_outliers'] and len(obs_list) > 0:
self._reject_outliers(obs_list)
obs_list.update_meta_data({'band_num': band})
coadd_obs_list.update_meta_data({'band_num': band})
return coadd_obs_list, obs_list
def _get_band_observation(self, band, mindex, icut):
"""
Get an Observation for a single band.
"""
meds = self.meds_list[band]
fname = self._get_meds_orig_filename(meds, mindex, icut)
im = self._get_meds_image(meds, mindex, icut)
wt, wt_us, seg = self._get_meds_weight(meds, mindex, icut)
jacob = self._get_jacobian(meds, mindex, icut)
# for the psf fitting code
wt = wt.clip(min=0.0)
psf_obs = self._get_psf_observation(band, mindex, icut, jacob)
obs = Observation(im, weight=wt.copy(), jacobian=jacob, psf=psf_obs)
if wt_us is not None:
obs.weight_us = wt_us.copy()
else:
obs.weight_us = None
obs.weight_raw = wt.copy()
obs.seg = seg
obs.filename = fname
return obs
def mbobs():
mbobs = MultiBandObsList()
for i in range(3):
ol = ObsList()
for j in range(4):
o = Observation(image=np.ones((32, 32)) * (j + 1),
weight=np.ones((32, 32)) * (j + 1),
jacobian=ngmix.DiagonalJacobian(scale=0.25,
row=0,
col=0))
ol.append(o)
mbobs.append(ol)
return mbobs
def get_psf_obs(*, rng, T=TPSF, model="turb", noise=1.0e-6):
dims = [25, 25]
cen = (np.array(dims) - 1.0) / 2.0
jacob = DiagonalJacobian(scale=PIXEL_SCALE, row=cen[0], col=cen[1])
gm = GMixModel([0.0, 0.0, 0.0, 0.0, T, 1.0], model)
im = gm.make_image(dims, jacobian=jacob)
im += rng.normal(scale=noise, size=im.shape)
weight = im*0 + 1.0/noise**2
return {
'obs': Observation(im, weight=weight, jacobian=jacob),
'gmix': gm,
}
def _get_psf_observation(self, band, mindex, icut, image_jacobian):
"""
Get an Observation representing the PSF and the "sigma"
from the psf model
"""
im, cen, sigma_pix, fname = self._get_psf_image(band, mindex, icut)
psf_jacobian = image_jacobian.copy()
psf_jacobian.set_cen(row=cen[0], col=cen[1])
psf_obs = Observation(im, jacobian=psf_jacobian)
psf_obs.filename = fname
# convert to sky coords
sigma_sky = sigma_pix * psf_jacobian.get_scale()
psf_obs.update_meta_data({'sigma_sky': sigma_sky})
psf_obs.update_meta_data({'Tguess': sigma_sky * sigma_sky})
psf_obs.update_meta_data({'psf_norm': im.sum()})
return psf_obs
def _get_band_observation(self, band, mindex, icut):
"""
Get an Observation for a single band.
"""
meds = self.meds_list[band]
bmask, skip = self._get_meds_bmask(band, meds, mindex, icut)
if skip:
return None
wt, wt_us, wt_raw, seg, skip = self._get_meds_weight(
band, meds, mindex, icut, bmask)
if skip:
return None
im = self._get_meds_image(meds, mindex, icut)
jacob = self._get_jacobian(band, mindex, icut)
if jacob is None:
return None
if self.conf['ignore_zero_images'] and 0.0 == im.sum():
print(" image all zero, skipping")
return None
psf_obs = self._get_psf_observation(band, mindex, icut, jacob)
obs = Observation(im,
weight=wt,
bmask=bmask,
jacobian=jacob,
psf=psf_obs)
if wt_us is not None:
obs.weight_us = wt_us
else:
obs.weight_us = None
obs.weight_raw = wt_raw
obs.seg = seg
nimage = self._get_meds_noise(meds, mindex, icut)
if nimage is not None:
obs.noise = nimage
if 'trim_image' in self.conf:
self._trim_obs(obs)
return obs
def _get_psf_observation(self, band, mindex, icut, image_jacobian):
"""
Get an Observation representing the PSF and the "sigma"
from the psf model
"""
im, cen, sigma_pix, fname = self._get_psf_image(band, mindex, icut)
psf_jacobian = image_jacobian.copy()
psf_jacobian.set_cen(cen[0], cen[1])
psf_obs = Observation(im, jacobian=psf_jacobian)
psf_obs.filename = fname
# convert to sky coords
sigma_sky = sigma_pix * psf_jacobian.get_scale()
psf_obs.update_meta_data({"sigma_sky": sigma_sky})
psf_obs.update_meta_data({"Tguess": sigma_sky * sigma_sky})
psf_obs.update_meta_data({"psf_norm": im.sum()})
return psf_obs
def _get_band_observation(self, band, mindex, icut):
"""
Get an Observation for a single band.
"""
meds=self.meds_list[band]
bmask,skip = self._get_meds_bmask(meds, mindex, icut)
if skip:
return None
wt,wt_us,wt_raw,seg,skip = self._get_meds_weight(meds, mindex, icut)
if skip:
return None
im = self._get_meds_image(meds, mindex, icut)
jacob = self._get_jacobian(meds, mindex, icut)
if self.conf['ignore_zero_images'] and 0.0==im.sum():
print(" image all zero, skipping")
return None
psf_obs = self._get_psf_observation(band, mindex, icut, jacob)
obs=Observation(im,
weight=wt,
bmask=bmask,
jacobian=jacob,
psf=psf_obs)
if wt_us is not None:
obs.weight_us = wt_us
else:
obs.weight_us = None
obs.weight_raw = wt_raw
obs.seg = seg
fname = self._get_meds_orig_filename(meds, mindex, icut)
obs.filename=fname
if 'trim_image' in self.conf:
self._trim_obs(obs)
return obs
def _make_mbobs(im, cen):
dx = cen[0] - int(cen[0] + 0.5)
dy = cen[1] - int(cen[1] + 0.5)
xlow = int(int(cen[0] + 0.5) - nstamp_cen)
ylow = int(int(cen[1] + 0.5) - nstamp_cen)
_im = im[ylow:ylow + nstamp, xlow:xlow + nstamp]
jac = DiagonalJacobian(
scale=wcs.scale,
x=nstamp_cen + dx,
y=nstamp_cen + dy,
)
obs = Observation(image=_im,
jacobian=jac,
psf=psf_obs,
weight=np.ones_like(_im),
noise=rng.normal(size=_im.shape) * noise)
mbobs = MultiBandObsList()
obsl = ObsList()
obsl.append(obs)
mbobs.append(obsl)
return mbobs
def make_ngmix_mbobslist(im, psf, wcs, cen1, cen2, noise, rng, nstamp=33):
nstamp_cen = (nstamp - 1) / 2
psf_im = psf.drawImage(nx=nstamp, ny=nstamp, wcs=wcs)
jac = DiagonalJacobian(scale=wcs.scale, x=nstamp_cen, y=nstamp_cen)
psf_obs = Observation(
image=psf_im.array,
jacobian=jac,
weight=np.ones_like(psf_im.array),
)
def _make_mbobs(im, cen):
dx = cen[0] - int(cen[0] + 0.5)
dy = cen[1] - int(cen[1] + 0.5)
xlow = int(int(cen[0] + 0.5) - nstamp_cen)
ylow = int(int(cen[1] + 0.5) - nstamp_cen)
_im = im[ylow:ylow + nstamp, xlow:xlow + nstamp]
jac = DiagonalJacobian(
scale=wcs.scale,
x=nstamp_cen + dx,
y=nstamp_cen + dy,
)
obs = Observation(image=_im,
jacobian=jac,
psf=psf_obs,
weight=np.ones_like(_im),
noise=rng.normal(size=_im.shape) * noise)
mbobs = MultiBandObsList()
obsl = ObsList()
obsl.append(obs)
mbobs.append(obsl)
return mbobs
return [
_make_mbobs(im, cen1),
_make_mbobs(im, cen2),
]
def test_ml_fitting_exp_obj_gauss_psf_smoke(g1_true, g2_true, wcs_g1, wcs_g2):
rng = np.random.RandomState(seed=10)
image_size = 33
cen = (image_size - 1) / 2
gs_wcs = galsim.ShearWCS(0.25, galsim.Shear(g1=wcs_g1,
g2=wcs_g2)).jacobian()
scale = np.sqrt(gs_wcs.pixelArea())
g_prior = ngmix.priors.GPriorBA(0.1)
cen_prior = ngmix.priors.CenPrior(0, 0, scale, scale)
T_prior = ngmix.priors.FlatPrior(0.01, 2)
F_prior = ngmix.priors.FlatPrior(1e-4, 1e9)
prior = ngmix.joint_prior.PriorSimpleSep(cen_prior, g_prior, T_prior,
F_prior)
gal = galsim.Exponential(half_light_radius=0.5).shear(
g1=g1_true, g2=g2_true).withFlux(400)
obj = galsim.Convolve([gal, galsim.Gaussian(fwhm=0.5)])
psf_im = galsim.Gaussian(fwhm=0.5).drawImage(nx=33,
ny=33,
wcs=gs_wcs,
method='no_pixel').array
psf_gmix = ngmix.gmix.make_gmix_model(
[0, 0, 0, 0, fwhm_to_T(0.5), 1], "gauss")
psf_obs = Observation(image=psf_im, gmix=psf_gmix)
im = obj.drawImage(nx=image_size,
ny=image_size,
wcs=gs_wcs,
method='no_pixel').array
noise = np.sqrt(np.sum(im**2)) / 1e16
wgt = np.ones_like(im) / noise**2
guess = np.ones(6) * 0.1
guess[0] = 0
guess[1] = 0
guess[2] = g1_true
guess[3] = g2_true
guess[4] = fwhm_to_T(0.5)
guess[5] = 400
g1arr = []
g2arr = []
farr = []
xarr = []
yarr = []
for _ in range(50):
shift = rng.uniform(low=-scale / 2, high=scale / 2, size=2)
xy = gs_wcs.toImage(galsim.PositionD(shift))
im = obj.shift(dx=shift[0],
dy=shift[1]).drawImage(nx=image_size,
ny=image_size,
wcs=gs_wcs,
method='no_pixel',
dtype=np.float64).array
jac = Jacobian(y=cen + xy.y,
x=cen + xy.x,
dudx=gs_wcs.dudx,
dudy=gs_wcs.dudy,
dvdx=gs_wcs.dvdx,
dvdy=gs_wcs.dvdy)
_im = im + (rng.normal(size=im.shape) * noise)
obs = Observation(image=_im, weight=wgt, jacobian=jac, psf=psf_obs)
fitter = LMSimple(obs, 'exp', prior=prior)
fitter.go(guess + rng.normal(size=6) * 0.01)
res = fitter.get_result()
if res['flags'] == 0:
_g1, _g2, _ = res['g'][0], res['g'][1], res['pars'][4]
g1arr.append(_g1)
g2arr.append(_g2)
farr.append(res['pars'][5])
xarr.append(res['pars'][1])
yarr.append(res['pars'][0])
g1 = np.mean(g1arr)
g2 = np.mean(g2arr)
gtol = 1e-5
assert np.abs(g1 - g1_true) < gtol
assert np.abs(g2 - g2_true) < gtol
xerr = np.std(xarr) / np.sqrt(len(xarr))
assert np.abs(np.mean(xarr)) < xerr * 5
yerr = np.std(yarr) / np.sqrt(len(yarr))
assert np.abs(np.mean(yarr)) < yerr * 5
def test_admom_smoke(g1_true, g2_true, wcs_g1, wcs_g2, weight_fac):
rng = np.random.RandomState(seed=100)
fwhm = 0.9
image_size = 107
cen = (image_size - 1) / 2
gs_wcs = galsim.ShearWCS(0.125, galsim.Shear(g1=wcs_g1,
g2=wcs_g2)).jacobian()
obj = galsim.Gaussian(fwhm=fwhm).shear(g1=g1_true,
g2=g2_true).withFlux(400)
im = obj.drawImage(nx=image_size,
ny=image_size,
wcs=gs_wcs,
method='no_pixel').array
noise = np.sqrt(np.sum(im**2)) / 1e18
wgt = np.ones_like(im) / noise**2
scale = np.sqrt(gs_wcs.pixelArea())
g1arr = []
g2arr = []
Tarr = []
for _ in range(50):
shift = rng.uniform(low=-scale / 2, high=scale / 2, size=2)
xy = gs_wcs.toImage(galsim.PositionD(shift))
im = obj.shift(dx=shift[0],
dy=shift[1]).drawImage(nx=image_size,
ny=image_size,
wcs=gs_wcs,
method='no_pixel',
dtype=np.float64).array
jac = Jacobian(y=cen + xy.y,
x=cen + xy.x,
dudx=gs_wcs.dudx,
dudy=gs_wcs.dudy,
dvdx=gs_wcs.dvdx,
dvdy=gs_wcs.dvdy)
_im = im + (rng.normal(size=im.shape) * noise)
obs = Observation(image=_im, weight=wgt, jacobian=jac)
# use a huge weight so that we get the raw moments back out
fitter = GaussMom(obs, fwhm * weight_fac, rng=rng)
try:
fitter.go()
res = fitter.get_result()
if res['flags'] == 0:
if weight_fac > 1:
_g1, _g2 = e1e2_to_g1g2(res['e'][0], res['e'][1])
else:
_g1, _g2 = res['e'][0], res['e'][1]
g1arr.append(_g1)
g2arr.append(_g2)
Tarr.append(res['pars'][4])
except GMixRangeError:
pass
g1 = np.mean(g1arr)
g2 = np.mean(g2arr)
gtol = 1e-9
assert np.abs(g1 - g1_true) < gtol, (g1,
np.std(g1arr) / np.sqrt(len(g1arr)))
assert np.abs(g2 - g2_true) < gtol, (g2,
np.std(g2arr) / np.sqrt(len(g2arr)))
# T test should only pass when the weight function is constant so
# weight_fac needs to be rally big
if g1_true == 0 and g2_true == 0 and weight_fac > 1:
T = np.mean(Tarr)
assert np.abs(T - fwhm_to_T(fwhm)) < 1e-6
def test_ml_fitting_exp_obj_gauss_psf_smoke(
g1_true, g2_true, wcs_g1, wcs_g2, fit_model):
rng = np.random.RandomState(seed=10)
image_size = 33
cen = (image_size - 1)/2
gs_wcs = galsim.ShearWCS(
0.25, galsim.Shear(g1=wcs_g1, g2=wcs_g2)).jacobian()
scale = np.sqrt(gs_wcs.pixelArea())
gal = galsim.Exponential(
half_light_radius=0.5
).shear(
g1=g1_true, g2=g2_true
).withFlux(
400,
)
obj = galsim.Convolve([gal, galsim.Gaussian(fwhm=0.5)])
psf_im = galsim.Gaussian(fwhm=0.5).drawImage(
nx=33, ny=33, wcs=gs_wcs, method='no_pixel').array
psf_gmix = ngmix.gmix.make_gmix_model(
[0, 0, 0, 0, fwhm_to_T(0.5), 1], "gauss")
psf_obs = Observation(
image=psf_im,
gmix=psf_gmix
)
im = obj.drawImage(
nx=image_size,
ny=image_size,
wcs=gs_wcs,
method='no_pixel',
).array
noise = np.sqrt(np.sum(im**2)) / 1e16
wgt = np.ones_like(im) / noise**2
prior = get_prior(fit_model=fit_model, rng=rng, scale=scale)
guess = prior.sample()
guess[0] = 0
guess[1] = 0
guess[2] = g1_true
guess[3] = g2_true
guess[4] = fwhm_to_T(0.5)
if fit_model == 'bd':
guess[5] = 1.0
guess[6] = 0.5
guess[7] = 400
elif fit_model == 'bdf':
guess[6] = 400
else:
guess[5] = 400
g1arr = []
g2arr = []
farr = []
xarr = []
yarr = []
fitter = Fitter(model=fit_model, prior=prior)
for _ in range(50):
shift = rng.uniform(low=-scale/2, high=scale/2, size=2)
xy = gs_wcs.toImage(galsim.PositionD(shift))
im = obj.shift(
dx=shift[0], dy=shift[1]
).drawImage(
nx=image_size,
ny=image_size,
wcs=gs_wcs,
method='no_pixel',
dtype=np.float64).array
jac = Jacobian(
y=cen + xy.y, x=cen + xy.x,
dudx=gs_wcs.dudx, dudy=gs_wcs.dudy,
dvdx=gs_wcs.dvdx, dvdy=gs_wcs.dvdy)
_im = im + (rng.normal(size=im.shape) * noise)
obs = Observation(
image=_im,
weight=wgt,
jacobian=jac,
psf=psf_obs)
res = fitter.go(
obs=obs, guess=guess + rng.normal(size=guess.size) * 0.01,
)
if res['flags'] == 0:
_g1, _g2, _ = res['g'][0], res['g'][1], res['pars'][4]
g1arr.append(_g1)
g2arr.append(_g2)
farr.append(res['pars'][5])
xarr.append(res['pars'][1])
yarr.append(res['pars'][0])
g1 = np.mean(g1arr)
g2 = np.mean(g2arr)
if fit_model == 'bd':
# bd fitting is highly degenerate, we don't recover the
# ellipticity with as much accuracy
gtol = 0.002
else:
gtol = 1.0e-5
assert np.abs(g1 - g1_true) < gtol
assert np.abs(g2 - g2_true) < gtol
xerr = np.std(xarr) / np.sqrt(len(xarr))
assert np.abs(np.mean(xarr)) < xerr * 5
yerr = np.std(yarr) / np.sqrt(len(yarr))
assert np.abs(np.mean(yarr)) < yerr * 5
def test_ml_max_fitting_gauss_smoke(g1_true, g2_true, wcs_g1, wcs_g2):
rng = np.random.RandomState(seed=42)
# allow some small relative bias due to approximate exp function
tol = 1.0e-5
flux = 400 # in galsim units
image_size = 33
cen = (image_size - 1) / 2
gs_wcs = galsim.ShearWCS(0.25, galsim.Shear(g1=wcs_g1,
g2=wcs_g2)).jacobian()
scale = np.sqrt(gs_wcs.pixelArea())
g_prior = ngmix.priors.GPriorBA(sigma=0.2, rng=rng)
cen_prior = ngmix.priors.CenPrior(
cen1=0,
cen2=0,
sigma1=scale,
sigma2=scale,
rng=rng,
)
T_prior = ngmix.priors.FlatPrior(minval=0.1, maxval=2, rng=rng)
F_prior = ngmix.priors.FlatPrior(minval=1e-4, maxval=1e9, rng=rng)
prior = ngmix.joint_prior.PriorSimpleSep(
cen_prior=cen_prior,
g_prior=g_prior,
T_prior=T_prior,
F_prior=F_prior,
)
obj = galsim.Gaussian(fwhm=0.9).shear(g1=g1_true,
g2=g2_true).withFlux(flux, )
im = obj.drawImage(nx=image_size,
ny=image_size,
wcs=gs_wcs,
method='no_pixel').array
noise = np.sqrt(np.sum(im**2)) / 1.e6
wgt = np.ones_like(im) / noise**2
g1arr = []
g2arr = []
Tarr = []
farr = []
xarr = []
yarr = []
for _ in range(10):
shift = rng.uniform(low=-scale / 2, high=scale / 2, size=2)
xy = gs_wcs.toImage(galsim.PositionD(shift))
im = obj.shift(dx=shift[0],
dy=shift[1]).drawImage(nx=image_size,
ny=image_size,
wcs=gs_wcs,
method='no_pixel',
dtype=np.float64).array
jac = Jacobian(
y=cen + xy.y,
x=cen + xy.x,
dudx=gs_wcs.dudx,
dudy=gs_wcs.dudy,
dvdx=gs_wcs.dvdx,
dvdy=gs_wcs.dvdy,
)
_im = im + rng.normal(size=im.shape, scale=noise)
obs = Observation(
image=_im,
weight=wgt,
jacobian=jac,
)
prior = None
fitter = Fitter(model='gauss', prior=prior)
guess = np.zeros(6)
guess[0:2] = rng.uniform(low=-0.1 * scale, high=0.1 * scale, size=2)
guess[2] = g1_true + rng.uniform(low=-0.01, high=0.01)
guess[3] = g2_true + rng.uniform(low=-0.01, high=0.01)
guess[4] = fwhm_to_T(0.9) * rng.uniform(low=0.99, high=1.01)
guess[5] = flux * rng.uniform(low=0.99, high=1.01)
res = fitter.go(obs=obs, guess=guess)
if res['flags'] == 0:
_g1, _g2, _T = res['g'][0], res['g'][1], res['pars'][4]
g1arr.append(_g1)
g2arr.append(_g2)
Tarr.append(_T)
farr.append(res['pars'][5])
xarr.append(res['pars'][1])
yarr.append(res['pars'][0])
assert len(g1arr) > 0
g1 = np.mean(g1arr)
g2 = np.mean(g2arr)
assert np.abs(g1 - g1_true) < tol
assert np.abs(g2 - g2_true) < tol
if g1_true == 0 and g2_true == 0:
T = np.mean(Tarr)
Ttrue = fwhm_to_T(0.9)
assert T / Ttrue - 1 < tol
fmn = np.mean(farr)
assert np.abs(fmn / flux - 1) < tol
xerr = np.std(xarr) / np.sqrt(len(xarr))
assert np.abs(np.mean(xarr)) < xerr * 5
yerr = np.std(yarr) / np.sqrt(len(yarr))
assert np.abs(np.mean(yarr)) < yerr * 5
def get_ngauss_obs(*, rng, ngauss, noise=0.0, with_psf=False, psf_model='turb'):
counts = 100.0
dims = [25, 25]
cen = (np.array(dims) - 1.0) / 2.0
jacob = DiagonalJacobian(scale=PIXEL_SCALE, row=cen[0], col=cen[1])
T_1 = 0.55 # arcsec**2
if with_psf:
T_1 = T_1 - TPSF
e1_1 = 0.1
e2_1 = 0.05
irr_1 = T_1 / 2.0 * (1 - e1_1)
irc_1 = T_1 / 2.0 * e2_1
icc_1 = T_1 / 2.0 * (1 + e1_1)
cen1 = [-3.25*PIXEL_SCALE, -3.25*PIXEL_SCALE]
if ngauss == 2:
frac1 = 0.4
frac2 = 0.6
cen2 = [3.0*PIXEL_SCALE, 0.5*PIXEL_SCALE]
T_2 = T_1/2
e1_1 = 0.1
e2_1 = 0.05
e1_2 = -0.2
e2_2 = -0.1
counts_1 = frac1 * counts
counts_2 = frac2 * counts
irr_1 = T_1 / 2.0 * (1 - e1_1)
irc_1 = T_1 / 2.0 * e2_1
icc_1 = T_1 / 2.0 * (1 + e1_1)
irr_2 = T_2 / 2.0 * (1 - e1_2)
irc_2 = T_2 / 2.0 * e2_2
icc_2 = T_2 / 2.0 * (1 + e1_2)
pars = [
counts_1,
cen1[0],
cen1[1],
irr_1,
irc_1,
icc_1,
counts_2,
cen2[0],
cen2[1],
irr_2,
irc_2,
icc_2,
]
elif ngauss == 1:
pars = [
counts,
cen1[0],
cen1[1],
irr_1,
irc_1,
icc_1,
]
gm = GMix(pars=pars)
if with_psf:
psf_ret = get_psf_obs(rng=rng, model=psf_model)
gmconv = gm.convolve(psf_ret['gmix'])
im0 = gmconv.make_image(dims, jacobian=jacob)
psf_obs = psf_ret['obs']
else:
im0 = gm.make_image(dims, jacobian=jacob)
psf_obs = None
im = im0 + rng.normal(size=im0.shape, scale=noise)
obs = Observation(im, jacobian=jacob, psf=psf_obs)
ret = {
'obs': obs,
'gmix': gm,
}
if with_psf:
ret['psf_gmix'] = psf_ret['gmix']
return ret
def get_model_obs(
*, rng, model,
noise=0.0,
psf_model='turb',
psf_noise=1.0e-6,
set_psf_gmix=False,
set_templates=False,
set_psf_templates=False,
nepoch=None, nband=None,
star=False,
):
if nband is not None:
do_mbobs = True
if nepoch is None:
nepoch = 1
else:
do_mbobs = False
nband = 1
if nepoch is not None:
do_obslist = True
else:
do_obslist = False
nepoch = 1
if star:
T = 0.0
else:
T = 0.27
g1 = 0.1
g2 = 0.05
flux = 100.0
off = 0.5
# not offset from the jacobian center
pars = [0.0, 0.0, g1, g2, T] + [flux]*nband
gm = GMixModel(pars[0:6], model)
mbobs = MultiBandObsList()
for iband in range(nband):
obslist = ObsList()
for i in range(nepoch):
off1_pix, off2_pix = rng.uniform(low=-off, high=off, size=2)
dims = [32, 32]
jcen = (np.array(dims) - 1.0) / 2.0
jacob = DiagonalJacobian(
scale=PIXEL_SCALE,
row=jcen[0] + off1_pix,
col=jcen[1] + off2_pix,
)
psf_ret = get_psf_obs(rng=rng, model=psf_model, noise=psf_noise)
if set_psf_gmix:
psf_ret['obs'].set_gmix(psf_ret['gmix'])
if set_psf_templates:
psf_ret['obs'].template = psf_ret['gmix'].make_image(
dims,
jacobian=jacob,
)
gmconv = gm.convolve(psf_ret['gmix'])
im0 = gmconv.make_image(dims, jacobian=jacob)
im = im0 + rng.normal(size=im0.shape, scale=noise)
if noise == 0.0:
weight = im*0 + 1.0/1.0e-12
else:
weight = im*0 + 1.0/noise**2
obs = Observation(
im,
weight=weight,
jacobian=jacob,
psf=psf_ret['obs'],
)
if set_templates:
obs.template = im0
obslist.append(obs)
mbobs.append(obslist)
ret = {
'gmix': gm,
'pars': pars,
'psf_data': psf_ret,
}
if do_mbobs:
ret['obs'] = mbobs
else:
obslist = mbobs[0]
if not do_obslist:
obs = obslist[0]
ret['obs'] = obs
else:
ret['obs'] = obslist
return ret
def test_admom_smoke(g1_true, g2_true, wcs_g1, wcs_g2):
rng = np.random.RandomState(seed=100)
fwhm = 0.9
image_size = 107
cen = (image_size - 1) / 2
gs_wcs = galsim.ShearWCS(0.125, galsim.Shear(g1=wcs_g1,
g2=wcs_g2)).jacobian()
obj = galsim.Gaussian(fwhm=fwhm).shear(g1=g1_true,
g2=g2_true).withFlux(400)
im = obj.drawImage(nx=image_size,
ny=image_size,
wcs=gs_wcs,
method='no_pixel').array
noise = np.sqrt(np.sum(im**2)) / 1e18
wgt = np.ones_like(im) / noise**2
scale = np.sqrt(gs_wcs.pixelArea())
g1arr = []
g2arr = []
Tarr = []
for _ in range(50):
shift = rng.uniform(low=-scale / 2, high=scale / 2, size=2)
xy = gs_wcs.toImage(galsim.PositionD(shift))
im = obj.shift(dx=shift[0],
dy=shift[1]).drawImage(nx=image_size,
ny=image_size,
wcs=gs_wcs,
method='no_pixel',
dtype=np.float64).array
jac = Jacobian(y=cen + xy.y,
x=cen + xy.x,
dudx=gs_wcs.dudx,
dudy=gs_wcs.dudy,
dvdx=gs_wcs.dvdx,
dvdy=gs_wcs.dvdy)
_im = im + (rng.normal(size=im.shape) * noise)
obs = Observation(image=_im, weight=wgt, jacobian=jac)
fitter = Admom(obs, rng=rng)
try:
fitter.go(fwhm_to_T(fwhm) + rng.normal() * 0.01)
res = fitter.get_result()
if res['flags'] == 0:
gm = fitter.get_gmix()
_g1, _g2, _T = gm.get_g1g2T()
g1arr.append(_g1)
g2arr.append(_g2)
Tarr.append(_T)
except GMixRangeError:
pass
g1 = np.mean(g1arr)
g2 = np.mean(g2arr)
gtol = 1e-6
assert np.abs(g1 - g1_true) < gtol, (g1,
np.std(g1arr) / np.sqrt(len(g1arr)))
assert np.abs(g2 - g2_true) < gtol, (g2,
np.std(g2arr) / np.sqrt(len(g2arr)))
if g1_true == 0 and g2_true == 0:
T = np.mean(Tarr)
assert np.abs(T - fwhm_to_T(fwhm)) < 1e-6
def test_admom_smoke(g1_true, g2_true, wcs_g1, wcs_g2):
rng = np.random.RandomState(seed=100)
fwhm = 0.9
image_size = 107
cen = (image_size - 1) / 2
gs_wcs = galsim.ShearWCS(0.125, galsim.Shear(g1=wcs_g1,
g2=wcs_g2)).jacobian()
obj = galsim.Gaussian(fwhm=fwhm).shear(g1=g1_true,
g2=g2_true).withFlux(400)
im = obj.drawImage(nx=image_size,
ny=image_size,
wcs=gs_wcs,
method='no_pixel').array
noise = np.sqrt(np.sum(im**2)) / 1e18
wgt = np.ones_like(im) / noise**2
scale = np.sqrt(gs_wcs.pixelArea())
g1arr = []
g2arr = []
Tarr = []
for _ in range(50):
shift = rng.uniform(low=-scale / 2, high=scale / 2, size=2)
xy = gs_wcs.toImage(galsim.PositionD(shift))
im = obj.shift(dx=shift[0], dy=shift[1]).drawImage(
nx=image_size,
ny=image_size,
wcs=gs_wcs,
method='no_pixel',
dtype=np.float64,
).array
jac = Jacobian(y=cen + xy.y,
x=cen + xy.x,
dudx=gs_wcs.dudx,
dudy=gs_wcs.dudy,
dvdx=gs_wcs.dvdx,
dvdy=gs_wcs.dvdy)
_im = im + (rng.normal(size=im.shape) * noise)
obs = Observation(image=_im, weight=wgt, jacobian=jac)
Tguess = fwhm_to_T(fwhm) + rng.normal() * 0.01
res = run_admom(obs=obs, guess=Tguess)
if res['flags'] == 0:
gm = res.get_gmix()
_g1, _g2, _T = gm.get_g1g2T()
g1arr.append(_g1)
g2arr.append(_g2)
Tarr.append(_T)
fim = res.make_image()
assert fim.shape == im.shape
res['flags'] = 5
with pytest.raises(RuntimeError):
res.make_image()
with pytest.raises(RuntimeError):
res.get_gmix()
g1 = np.mean(g1arr)
g2 = np.mean(g2arr)
gtol = 1.5e-6
assert np.abs(g1 - g1_true) < gtol, (g1,
np.std(g1arr) / np.sqrt(len(g1arr)))
assert np.abs(g2 - g2_true) < gtol, (g2,
np.std(g2arr) / np.sqrt(len(g2arr)))
if g1_true == 0 and g2_true == 0:
T = np.mean(Tarr)
assert np.abs(T - fwhm_to_T(fwhm)) < 1e-6
with pytest.raises(ValueError):
_ = run_admom(None, None)
# cover some branches
tres = copy.deepcopy(res)
tres['flags'] = 0
tres['sums_cov'][:, :] = np.nan
tres = ngmix.admom.admom.get_result(tres)
assert tres['e1err'] == 9999.0
tres = copy.deepcopy(res)
tres['flags'] = 0
tres['pars'][4] = -1
tres = ngmix.admom.admom.get_result(tres)
assert tres['flags'] == 0x8
def test_ml_fitting_gauss_smoke(g1_true, g2_true, wcs_g1, wcs_g2):
rng = np.random.RandomState(seed=42)
# allow some small relative bias due to approximate exp function
tol = 1.0e-5
image_size = 33
cen = (image_size - 1) / 2
gs_wcs = galsim.ShearWCS(0.25, galsim.Shear(g1=wcs_g1,
g2=wcs_g2)).jacobian()
scale = np.sqrt(gs_wcs.pixelArea())
g_prior = ngmix.priors.GPriorBA(0.2)
cen_prior = ngmix.priors.CenPrior(0, 0, scale, scale)
T_prior = ngmix.priors.FlatPrior(0.1, 2)
F_prior = ngmix.priors.FlatPrior(1e-4, 1e9)
prior = ngmix.joint_prior.PriorSimpleSep(cen_prior, g_prior, T_prior,
F_prior)
obj = galsim.Gaussian(fwhm=0.9).shear(g1=g1_true, g2=g2_true).withFlux(400)
im = obj.drawImage(nx=image_size,
ny=image_size,
wcs=gs_wcs,
method='no_pixel').array
noise = np.sqrt(np.sum(im**2)) / 1e16
wgt = np.ones_like(im) / noise**2
g1arr = []
g2arr = []
Tarr = []
farr = []
xarr = []
yarr = []
for _ in range(100):
shift = rng.uniform(low=-scale / 2, high=scale / 2, size=2)
xy = gs_wcs.toImage(galsim.PositionD(shift))
im = obj.shift(dx=shift[0],
dy=shift[1]).drawImage(nx=image_size,
ny=image_size,
wcs=gs_wcs,
method='no_pixel',
dtype=np.float64).array
jac = Jacobian(y=cen + xy.y,
x=cen + xy.x,
dudx=gs_wcs.dudx,
dudy=gs_wcs.dudy,
dvdx=gs_wcs.dvdx,
dvdy=gs_wcs.dvdy)
_im = im + (rng.normal(size=im.shape) * noise)
obs = Observation(image=_im, weight=wgt, jacobian=jac)
fitter = LMSimple(obs, 'gauss', prior=prior)
guess = np.ones(6) * 0.1
guess[0] = 0
guess[1] = 0
guess[2] = g1_true
guess[3] = g2_true
guess[4] = fwhm_to_T(0.9)
guess[5] = 400 * scale * scale
fitter.go(guess + rng.normal(size=6) * 0.01)
res = fitter.get_result()
if res['flags'] == 0:
_g1, _g2, _T = res['g'][0], res['g'][1], res['pars'][4]
g1arr.append(_g1)
g2arr.append(_g2)
Tarr.append(_T)
farr.append(res['pars'][5])
xarr.append(res['pars'][1])
yarr.append(res['pars'][0])
g1 = np.mean(g1arr)
g2 = np.mean(g2arr)
assert np.abs(g1 - g1_true) < tol
assert np.abs(g2 - g2_true) < tol
if g1_true == 0 and g2_true == 0:
T = np.mean(Tarr)
Ttrue = fwhm_to_T(0.9)
assert T / Ttrue - 1 < tol
fmn = np.mean(farr) / scale / scale
assert np.abs(fmn / 400 - 1) < tol
xerr = np.std(xarr) / np.sqrt(len(xarr))
assert np.abs(np.mean(xarr)) < xerr * 5
yerr = np.std(yarr) / np.sqrt(len(yarr))
assert np.abs(np.mean(yarr)) < yerr * 5
