GalSim version 2.3.0 released

[Mike Jarvis]

Hi all,

We are pleased to announce a new version of GalSim has been released, version 2.3.0.  It is available on pip now and should be on conda-forge sometime next week, probably.

The change log is available online.  Or see below if you prefer reading it in email.

The full documentation for version 2.3 is here.

Peace,

Mike Jarvis

on behalf of all the GalSim Developers

Changes from v2.2 to v2.3

This release mostly focuses on improvements needed for LSST Rubin simulations for the Dark Energy Science Collaboration and ones needed for Roman Space Telescope simulations for the Roman HLS SIT. These mostly involve adding features to the config layer processing that are relevant to these two telescopes as well as some new features to add additional realism to the simulations.

With this release, we are no longer supporting Python 3.5. And we have upgraded Python 3.8 and 3.9 to primary supported platforms, which now include 2.7, 3.6, 3.7, 3.8, 3.9.

Warning

This will be the last GalSim release to support the following legacy options:

• Python 2.7
• TMV for matrices in C++
• Boost Python for C++ bindings
• SCons builds

If any of these will cause hardship for you, please open an issue, so we can help you find mitigation strategies.

A full list of changes in this release are below. The numbers in parentheses are GalSim issue or pull request numbers where the change was implemented.

cf. https://github.com/GalSim-developers/GalSim/milestone/20?closed=1

Dependency Changes

• Removed future as a dependency. (#1082)
• Removed eigency as an optional way to get eigen dependency. Now it will download the eigen code directly if it cannot find eigen installed on your system. (#1086)

API Changes

• Deprecated the rng parameter to WavelengthSampler and FRatioAngles constructors. Now they will use the rng that you provide to drawImage or whatever else you are using to apply these photon operators. (#540)
• Deprecated withOrigin method for non-local WCS types in favor of the new method shiftOrigin. This has the same functionality, but the name is more in line with the actual action of the function. For local WCS types, shiftOrigin is equivalent to withOrigin, which still exists as a valid name for this action. (#1073)
• The numerical details of the Kolmogorov class were updated to use a more precise version of a constant from Racine (1996). Technically this changes the definition of the Kolmogorov profile at the 6th decimal place. So if you carefully tuned an r0 value to 6 decimal places for some purpose, this might break that. (#1084)
• Deprecated galsim.wfirst module. Now called galsim.roman. (#1088)
• Changed the default ii_pad_factor for PhaseScreenPSF and OpticalPSF to 1.5. The old default of 4.0 (matching the InterpolatedImage default) is not necessary for objects that will not be significantly sheared, which PSF objects typically aren’t. For almost all use cases, this will be simply a performance improvement, but if you need the higher ii_pad_factor for some reason, you will now need to manually specify it. (#1089)
• Deprecated the high_accuracy and approximate_struts parameters for the roman.getPSF function. You should now use pupil_bin and gsparams to effect similar adjustments to the default (which is now much better than before in terms of both speed and accuracy). (#1089)
• Deprecated the surface_ops parameter to GSObject.drawImage, switching to the name photon_ops, since these don’t have to be something that happens at the surface of the sensor. (#1093)
• Added logger option to some config functions and methods. If you are using custom Image or Output types, you may need to add a logger=None optional parameter to some method signatures. (#1095)
• Deprecated galsim.integ.trapz. You should use galsim.integ.int1d instead, which is almost always more efficient. (#1098)
• Deprecated galsim.integ.midpt. You should use np.trapz or galsim.trapz instead, which are almost equivalent, but slightly more accurate. (#1098)
• Changed the convention for the f array passed to the LookupTable2D constructor to be the transpose of what it was. This is arguably a bug fix, since the old convention was the opposite of every other array used in GalSim that reprented (x,y) positions. But if you have been using LookupTable2D, you will need to add .T (or remove it) from the f argument you are passing to the class. (#1103)
• Changed the behavior of PhaseScreenPsf, OpticalPsf, and ChromaticOpticalPsf by adding the kwarg fft_sign, which controls the sign in the exponent of the Fourier kernel using in the Fourier optics PSF equation. The new default value of ‘+’ produces a profile that is 180 degrees rotated compared to the former behavior. To revive the former behavior, set fft_sign='-'. (#1104)

Config Updates

• Added ability to draw chromatic objects with config files. (#510)
• Added demo12.yaml and demo13.yaml to the demo suite. (#510, #1121)
• Fixed a few issues with parsing truth catalog items and, in general, places where a Current item might be used before it was parsed. (#1083)
• Added value-type-specific type names like Random_float, Random_Angle, etc. to help in some cases where the config processing cannot know what value type to use for further processing. (#1083)
• Fixed a subtle issue in Eval string processing if one Current item is a shorter version of another one. e.g. @gal and @gal.index. It had been the case that the longer one might not be parsed properly. (#1083)
• Added photon_ops and sensor as options in the stamp processing. (#1093)
• Removed the _nobjects_only mechanism from input objects. If this pattern was useful for a custom input type, you should switch to using lazy properties to delay the loading of data other than what is needed for determining the number of objects. See Catalog for an example of how to do this. (#1095)
• Allowed Eval fields to use any modules that are listed in the top-level modules field. (#1121)
• Added Roman config types: RomanSCA, RomanBandpass, and RomanPSF. (#1121)

New Features

• Added atRedshift method for ChromaticObject. (#510)
• Added galsim.utilities.pickle_shared() context, which lets the shared portion of an AtmosphericScreen be included in the pickle. This allows the pickles to be recovered after writing to disk and reading back in, which otherwise would not work. (#1057)
• Added force_stepk option to VonKarman. (#1059)
• Added Refraction and FocusDepth photon ops. (#1065, #1069)
• Updated LSST sensor files to match new lab measurements and use improved Poisson code calculations. (#1077, #1081)
• Added makePhot method of GSObject. (#1078)
• Made it easier to set specific GSParams parameters using the syntax (e.g.) obj.withGSParams(folding_threshold=1.e-3) to just change that one value and keep any other non-default parameters the same. (#1089)
• Added a new pupil_bin option to the Roman getPSF function. This controls the resolution of the pupil plane mask, which involves a trade-off between speed and accuracy of the PSF rendering. (#1089)
• Added FittedSIPWCS to fit a WCS from a list of image and celestial coordinates. (#1092)
• Extended GSFitsWCS to support -SIP distortions for non-TAN WCSs. (#1092)
• Added wcs option to Roman getPSF function to more easily get the right PSF in world coordinates for a particular observation. (#1094)
• Added Position.shear method. (#1090)
• Added LookupTable.integrate and LookupTable.integrate_product, along with galsim.trapz as a drop in replacement for numpy.trapz, which is often somewhat faster. (#1098)
• Added galsim.integ.hankel function. (#1099)
• Added galsim.bessel.jv_root function. (#1099)
• Added support for TPV WCS files with order > 3. (#1101)
• Added galsim.UserScreen for arbitrary user-supplied phase screens (#1102)
• Added option to emit WCS warnings when reading a file via galsim.fits.read e.g. if the WCS defaulted to a PixelScale, or it reverted to an approximate AffineTransformation rather than the correct WCS. (#1120)
• Added area and exptime parameters to COSMOSCatalog constructor to make it easier to rescale the fluxes to a different telescope than HST. (#1121)
• Added zernike.describe_zernike to construct an algebraic string describing circular Zernike terms in the Cartesian basis. (#1104)

Performance Improvements

• Implemented Transformation._drawReal and _drawKImage in python to provide hooks for performance improvements in user code when these are used in tight loops. (#934)
• Sped up the draw routines for InterpolatedImage. (#935)
• Improved the rendering of Roman PSFs to always show 12 diffraction spikes (rather than 6 in the now-deprecated approximate_struts mode), remove an FFT artifact in the exact pupil plane mode, and significantly speed up all PSF renderings. (#1089)
• Sped up GSFitsWCS.radecToxy for SIP and PV distorted WCSs by ~20x. (#1092)
• Various speed improvements in config processing. (#1095, #1098)
• Sped up SED.calculateFlux and a few other SED and Bandpass calculations by switching to LookupTable.integrate_product for the implementation of the integrals. (#1098)
• Sped up the Hankel transforms several classes use for computing either the k-space values (e.g. Sersic) or real-space values (e.g. Kolmogorov). (#1099)
• Sped up Zernike arithmetic for the case where you just want to evaluate a resulting Zernike series without knowing its coefficients. (#1124)
• Removed some small bits of overhead in some “leading underscore” methods (e.g. _drawReal, _Transform, _shift, etc.) to make them faster. (#1126)

Bug Fixes

• Fixed horner and horner2d when inputs are complex. (#1054)
• Fixed VonKarman integration to be more reliable for various combinations of (r0, L0, lam). (#1058)
• Fixed minor bug in repr of OpticalPSF class. (#1061)
• Fixed bug in RandomKnots when multiplied by an SED. (#1064)
• Fixed bug that galsim.fits.writeMulti didn’t properly write the header information in each hdu. (#1091)
• Improved the accuracy of stepk for Kolmogorov profiles, especially when folding_threshold is very small. (#1110)