Skip to content

Latest commit

 

History

History
349 lines (254 loc) · 12.2 KB

CHANGES.md

File metadata and controls

349 lines (254 loc) · 12.2 KB
Raw

v2.3.2 (not yet released)

Bug Fixes

- Fixed bug when moments are used in guesser and size is bad.
  Only affected rng such as np.random.default_rng that checks
  range for uniform

v2.3.1

new features

- Add calculation of "rho4" to the adaptive moments code.

Bug Fixes

- The metacal psf reconvolution method 'gauss' (which uses MetacalGaussPSF)
  now works with a complex WCS

v2.3.0

new features

  • Caching in pre-psf moments and metacal is now optional with an API to turn it on. Default is off.
  • Pixel in pre-PSF moments is not deconvolved when doing PSFs.

v2.2.1

New Features

  • Added new guesser BDFPSFFluxGuesser, makes a flux guess based on the PSF flux

Performance

  • Added O(N) phase shift computation for pre-psf moments

Misc

  • ignore NumbaExperimentalFeatureWarning being spewed by numba. We can ignore these, as the features are of order ten years old from the time of writing

v2.2.0

new features

  • Added function to regularize moments results.
  • Added numba to pre-PSF moments to reduce runtime.

v2.1.0

bug fixes

- Fixed 1/area normalization for all moments fields in `GaussMom` results.

new features

- Added `fwhm_smooth` keyword to pre-PSF moments routines to allow for extra
  smoothing of the profile before the moments are measured.
- Added caching of FFTs in metacal and pre-PSF moment rountines.
- Added moments weight function normalization.
- Changed `mom*` fields in the moments outputs to `sums*`.
- Added `sums_norm` field to adaptive moments outputs.

v2.0.6

bug fixes

- Fixed a bug where the `kind` attribute was not set for the `PrePSFMom` fitter.

v2.0.5

new features

- implemented check for equality, copy() method, copy, and deepcopy
  for Gmix objects, jacobians, and Observation/ObsList/MultiBandObsList
- added get_data() method for jacobian objects
- added apodization to the pre-PSF moments to help prevent FFT artifacts

v2.0.4

new features

- util.get_ratio_error and get_ratio_var support array arguments
- added new function ngmix.gmix.gmix_concat to concatenate
  GMix objects into a combined GMix
- It is no longer required to send a random number generator
  to the constructors for MetacalGaussPSF and MetacalAnalyticPSF.  If one
  is sent it is still used to add a small amount of noise.
- The metacal codes now use a copy() of the input observations when
  creating their output.  This passes along all the internal data.

bug fixes

- Changes to ensure the "flux" result from the `AdmomFitter` is normalized
  the same as those from the other moments fitters.
- In MetacalFitGaussPSF when using adaptive moments to get the psf
  model, be careful to use the fitted parameters rather than going through
  an intermediate gmix, which converts e1, e2 to g1, g2 and can fail in
  rare cases.
- Some AdmomFitter failures were not being flagged properly.

v2.0.3

new features

- All moments codes now produce similar output dictionaries and flags.
- The `PrePSFGaussMom` class has been renamed to `PGaussMom`. The old name still
  exists for backwards compatibility but will be deprecated and removed at
  some future date.
- The moments codes and `gaussap.py` codes now use np.nan for missing values.
- The flag bits have been normalized across the code base to now all reside in
  `ngmix.flags`. A new function to convert the flags to a descriptive string is
  located there as well.
- A new function exists to convert raw moments to shape results:
  `ngmix.moments.make_mom_result`
- The moments fitters now have a `kind` attribute to help downstream code
  process them uniformly.
- Added slots for position moments in moments output for moments fitters.

bug fixes

- Ensure only square images are accepted for pre psf moments codes
  KSigmaMom and PrePSFGaussMom

v2.0.2

new features

  • Added PrePSFGaussMom fitter for computing pre-PSF moments in Gaussian apertures
  • Test suite now runs in parallel using pytest-xdist
  • Added flux_flags and T_flags fields to KSigmaMom and PrePSFGaussMom fitters for using just those quantities.
  • Added FFTRangeError which is now raised if KSigmaMom or PrePSFGaussMom are using inconsistent FFT sizes given the kernel size.
  • add cenonly option to adaptive moments, using a fixed gaussian covariance with only the center varying. Added ngmix.admom.find_cen_admom function for finding the center.

bug fixes

  • Fixed bug identifying that adaptive moments had reached maximum number of iterations

v2.0.1

new features

  • em fitting now supports fixed covariance matrices using the EMFitterFixCov, or equivalently the fixcov=True keyword to the run_em convenience function

bug fixes

  • fixed bug with sheared psfs in metacal, negative sheared psfs were not being cached.
  • bug in psf guesser when using guess_from_moms that caused infinite recursion

v2.0.0

breaking API changes

  • ngmix now evaluates models in flux units rather than surface brightness. Users whose images are in flux units previously had to include extra factors of pixel scale squared to get the right units, this is no longer the case.
  • The fitters no longer take the observation in the constructor. All fitting/measurement conform to a common interface. The classes have a go method that takes the observation and possibly a guess.
  • The bootstrap and runner code has been completely replaced by a new api. The bootstrappers are now generic, wrapping runners that are themselves generic wrappers for fitters and guessers. Thus the bootstrap/runner code is completely generic and can work with any of the fitters/guessers.
  • The guessers now optionally take the observation as argument when called; this facilitates the cases where a guess can be made based on the observations.
  • The em codes take GMix objects as guesses, a simple T guess is no longer supported
  • The em codes now work in image flux units rather than normalizing the mixture
  • all PDFs/priors sample methods take the option nrand=. Some had previously taken n=
  • fitters renamed:
    • LMSimple -> Fitter
    • LMCoellip -> CoellipFitter
    • TemplateFluxFitter -> PSFFluxFitter
    • GalsimSimple -> GalsimFitter
    • SpergelFitter -> GalsimSpergelFitter
    • MoffatFitter -> GalsimMoffatFitter
    • GalsimTemplateFluxFitter -> GalsimPSFFluxFitter
    • GMixEM -> EMFitter
    • Admom -> AdmomFitter
  • Copies of observations now propagate the values of store_pixels and ignore_zero_weight.

bug fixes

  • metacal was reconvolving by the pixel twice, which resulted in larger reconvolved PSFs and thus somewhat lower, but still accurate, response than the pixel was included once. This did not cause any bias.
  • fixed computation of Jacobian center for ngmix.KObservations with mixed even-odd dimensions
  • the total shape amplitude g is now reset properly when using the Shape.set_g1g2 method
  • fixed large rounding errors in some shape conversions in ngmix.shape
  • fixed bug in ngmix.shape.g1g2_to_eta1eta2 in handling numpy array inputs
  • fixed bug in ngmix.shape.e1e2_to_eta1eta2 in handling numpy array inputs
  • fixed bug in ngmix.priors.GPriorBA.get_fdiff in handling array inputs
  • fixed a bug in ngmix.priors.TwoSidedErf.get_fdiff in handling array inputs
  • fixed bug in TruncatedSimpleGauss2D where the truncation radius was applied about zero in some methods and not about the center
  • fixed flux unit scaling in metadetect regression tests

new features

  • All fitters and measurement classes adhere to a common interface. fitter.go(obs=obs) or fitter.go(obs=obs, guess=guess). The obs is no longer taken in the constructor so the same object can be reused in the runners/bootstrappers for multiple measurements.
  • Replaced boostrapper/runner classes with generic classes/functions that expect fitters/measurements and runners to provide common interfaces For fitters fitter.go(obs=obs) or fitter.go(obs=obs, guess=guess) For runners runner.go(obs=obs). New classes are bootstrap.Bootstrapper runners.Runner, runners.PSFRunner
  • Expanded test suite and improved documentation for all modules
  • introduced NGmixBaseException as parent class for all ngmix-specific exceptions
  • New guessers TPSFFluxGuesser, TPSFFluxAndPriorGuesser, GmixPSFGuesser, SimplePSFGuesser, CoellipPSFGuesser
  • New joint prior PriorGalsimSimpleSep
  • Added guessers TPSFFluxGuesser, TPSFFluxAndPriorGuesser, GMixPSFGuesser, SimplePSFGuesser, CoellipPSFGuesser,
  • New specialized EM fitting codes added:
    • GMixEMFixedCen: fit mixtures with fixed centers for each component
    • GMixEMFluxOnly: fit mixtures allowing only the fluxes to vary
  • some refactoring of like modules into sub-packages.
  • Added mfrac attribute to observations to hold masked fraction images.
  • Added Python @property functions for store_pixels and ignore_zero_weight for ngmix.Observation.
  • setup.py now gets the version from the package.
  • Added new KSigmaMom fitter to measure pre-PSF weighted moments using the ksigma kernel from Bernstein et al.

deprecated/removed

  • The LMSimple class now supports the "bdf" and "bd" models and has been renamed to LM. The LMSimple is now an alias for LM and is considered deprecated. The LMBDF and LMBD classes have been removed.
  • ngmix.lensfit has been removed in v2.0.0
  • all dbyg1_*, dbyg2_*, dlnbyg1_*, dlnbyg2_*, and get_pqr* methods, along with the code used to test them, have been removed from the classes in ngmix.priors
  • removed travis-ci in favor of GitHub Actions
  • the optional rng keyword for ngmix.priors.srandu has been removed in favor of a required keyword
  • removed priors ZDisk2DErf, ZAnnulus, UDisk2DCut, TruncatedStudentPolar, TruncatedGaussianPolar, Student, Student2D, MultivariateLogNormal, MVNMom, TruncatedSimpleGauss2D, GPriorGreat3Exp, GPriorGreatDES, GPriorGreatDES2, GPriorGreatDESNoExp, GPriorM, GPriorMErf, GPriorMErf2, FlatEtaPrior, BFrac, TFluxPriorCosmosExp, TFluxPriorCosmosDev, GPriorCosmosSersicSpline
  • Removed all the old bootstrapper and runner classes; replaced with generic Bootstrapper, Runner, PSFRuner, and MetacalBootstrapper, and associated functions
  • removed GMixBDF class. The "bdf" is now supported by the GMixModel class
  • removed the roundify and stats modules
  • removed the "prepix" option for metacal
  • removed guessers RoundParsGuesser, MomGuesser,
  • switched from print to logging in many modules
  • removed simplex.py no longer needed, we just use the scipy simplex fitter

v1.3.9

bug fixes

  • metacal was reconvolving by the pixel twice, which resulted in larger reconvolved PSFs and thus somewhat lower, but still accurate, response than the pixel was included once. This did not cause any bias.

v1.3.8

bug fixes

  • fix bug not updating pixel array when adding noise in metacal

new features

  • added writeable context for observations, now returned references for observation images etc. are read only unless in the writeable context
  • jacobian getter returns new Jacobian with readonly view, rather than a copy
  • added more unit tests

v1.3.7

new features

  • Add option to not store pixels

v1.3.6

bug fixes

  • fixed bug in T fraction sum for dev profiles
  • fixed bugs for the full bulge+disk fitter

new features

  • added order 5 fast exponential to fastexp.py which is exported as fexp. This has satisfactory accuracy but is much faster than expd in some real world scenarios. Modified the tests accordinly.
  • added a Gaussian moments fitter
  • added 5 gaussian coellip fitting in the coellip psf fitter

v1.3.5

bug fixes

  • better fast exponential function approximation
  • bug in gaussian aperture flux calculation for cm

new features

  • unit tests and travis CI for core APIs

v1.3.4

new features

  • analytic Gaussian aperture fluxes

v1.3.3

bug fixes

  • fixed bug in BDF Gaussian aperture fluxes

v1.3.2

bug fixes

  • Use psf_cutout_row and col for center in psf obs for meds reader