Revision History
Changes from version 1.6 to 1.7
This version mostly adds a bunch of things we needed to enable running Piff on Roman
Space Telescope images using an optical model based on the galSim.roman PSF model.
The new PSF type is called RomanOptics. See the RomanOpticsPSF docstring for details
about the new type. You can also see some example configurations that use it in the
devel/roman directory.
Dependency Changes
Added
pyarrowdependency for reading Parquet input catalogs. It is only required if you attempt to read a parquet file. (#188)Removed support for Python 3.8, 3.9.
Added support for Python 3.14, 3.15.
API Changes
Deprecated
input.nstarsin favor ofselect.nstars, which now works across all input files. The old option is deprecated and retains the per-image behavior. (#190)Renamed
select.max_snrtoselect.max_snr_weightto clarify that this caps a star’s effective fitting weight, rather than rejecting bright stars. The old name is still works but is deprecated. (#195)Deprecated the use of strings to mean FITS-header keywords for
noise,gain,saturandskyin theinputsection. You should now use the newImageHeaderValuetype instead. (#207)Various minor signature changes to Python functions. If you are using the Python interface rather than config files, you may need to check the doc strings for changes to return values or parameter names.
Output file changes
Added the number of fitting iterations,
niter, to the saved fit diagnostics. (#196)Added the observing bandpass when one is supplied. (#194)
Star SEDs are saved as file names and reconstruction metadata, rather than embedding the full spectra. If you do something that needs the SEDs after reading the psf from a file, those SED files must be accessible. (#194, #198)
Renamed the WCS table’s
nchunkscolumn towcs_nchunks. Files with the old name should continue to work properly. (#198)
New features
Added
input.badpix_zerosto mask zero-valued science pixels. (#185)Added support for reading the noise variance from the FITS header. (#186)
Added an option to compute the sky level from the median of the input image. Use
input.sky: median. (#187)Added support for Parquet input catalogs, using
pyarrow. (#188)Added the ability to read multiple image HDUs from a single FITS file to be processed as separate images. (#189)
Changed
nstarsto mean the total number of stars to use across all input images when there are more than one, rather than each image separately. Also moved this toselectwhere it fits more naturally. (#190)Added
raw_snrattribute for stars to keep track of their real snr, rather than the possibly adjusted value insnrdue toselect.max_snr_weight. (#190)Added the ability to compute star
propertiesderived from one or more columns in the input catalog. (#191)Added
Centroidoutlier rejection with amax_offsetthreshold, and allowoutliersto be a list combining multiple rejection criteria. (#192)Added
select.min_septo reject objects with a nearby neighbor. (#193)Added
input.bandpassfor specifying the bandpass used for the observations. (#194)Added support stellar SEDs in the input field for chromatic fitting. You can either specify a single spectrum to use for all stars (using
input.sed_file_name) or specify a specific spectrum for each star separately (usinginput.sed_col). There are additional parameters that can control how these files are used. See docs forInputFilesfor more details. (#194, #198)Added the
RomanOpticsPSFclass, based on GalSim’s Roman optical model, with fitted Zernike aberrations and Gaussian priors. (#195, #196, #197, #198)Changed
StarImagesplots to include chip number and image coordinates for each star. (#196)Expanded support for GalSim-style config value types in the config file, including
EvalandCurrentexpressions and their$and@shortcuts. (#207)
Bug fixes
Ensure input stamps use floating-point data, so sky subtraction and other processing work with integer-valued source images. (#187)
Fixed composite PSF serialization when components have parameter-variance arrays of different sizes. (#195)
Reject failed HSM measurements when using
hsm_size_rejectormax_pixel_cut. (#196)Fixed
StarImageswhen fewer eligible stars are available than requested. (#196)Fixed GSObject fitting to avoid nonpositive sizes. (#200)
Added checks for excessively large FFTs when fitting and reject them rather than risking huge memory allocations or crashes. (#205)
Fixed the
TwoDHistmodel-size panel, which previously repeated the measured star sizes instead of showing model sizes. (#205)Added warning about zero-S/N candidates when no
min_snrcut is set, with an additional suggestion to supply noise information when it is missing. (#200)Fixed chi-squared outlier rejection for already-flagged stars. (#208)