Published October 7, 2021
| Version 4.14.0
Software
Open
sherpa/sherpa: Sherpa 4.14.0
Creators
- 1. Smithsonian Astrophysical Observatory
- 2. MIT
- 3. Center for Astrophysics | Harvard & Smithsonian
- 4. HeidelbergCement
- 5. Caltech/IPAC
- 6. @spacetelescope
- 7. @gitlabhq
Description
Sherpa 4.14.0
This release of Sherpa includes various documentation updates, bug fixes, enhancements, and infrastructure changes.
- enhancements:
- filtering and grouping area for binned (1D) spectral data has been improved with changes to the default behavior and many bug fixes resulting in changes to the statistics, degrees-of-freedom and energy flux in comparison to the previous version for the same data with the same filter.
- updates to allow users to change the hard limits of XSPEC model parameters
- the sample_flux routine now returns correct information for the clip column
- documentation changes:
- improved PHA simulation documentation
- improved Filtering and grouping of PHA data documentation
- added sherpa.image module documentation
- added section on running tests to developer docs
- Infrastructure Changes:
- updates to support Apple ARM
- update to support Xspec version 12.12
- update fftw from version 3.3.8 to 3.3.9
- clean up of compiler and sphinx warnings
- changes to support gcc 9.3.0 in conda defaults
- updates to support python 3.9 including readline 8.1 upgrade, numpy minimum 1.19 (numpy 1.18 minimum for python 3.7/8)
- test infrastructure clean up and updates
- bug fixes:
- updates to fix several 'unable to parse region string: None' errors
- fix issue where save_all() of a loaded image with no region filter would fail on reload
- fixed issue with plot_model() being called before notice or ignore could lead to filters not getting applied
- fix to error out instead of crash when grouping data using an unsupported method
Update the estimation of FWHM for 1D profiles, and hence the guess method for
Gauss1D and related routines. The 2D models use the same routine so see these
changes.
The fake_pha command now treats id=None as the default id. This addresses #1064.
The sample_flux routine now returns correct information for the clip column
(that is, it matches the clipping done by this routine). There may be changes to
the reported error ranges because of this change.
Add keywords to the default output in the _repr_html_ for some X-ray classes.
Allow XSPEC ETABLE table model files to be read by load_xstable_model by
setting the etable parameter.
Add additional test cases for source plots with PHA data to check out
rarely-used combinations.
Improve some corner cases for filtering PHA data, including fixing #921 (channel
bounds are integers, not half-integers).
Avoid running a test we know to fail for an uncommon combination of options.
The previous example implied that a spectrum may contain both positive and
negative orders in the same spectrum. That can happen only after coadding -
and in that case one would co-add the arfs as well.
Remove sherpa.astro.models.MultiResponseSumModel which did nothing and seems
to be unused (except in a test that explicitly skip this class).
Change the wording of an error message. Potentially breaking if someone tests
for the wording of the error message in their code, but certainly more correct.
The old docstring claims that the background is just scaled and not simulated,
but there is definitely a call to poisson_noise for the background in the code.
NumPy 1.20 notes that numpy.bool, numpy.int, and numpy.float will be removed
and users should just use bool, int, and float instead, so these symbols have
been changed. This is a follow on to #1092
clean up references to master with main
Add basic testing of the integrate1d model class and improve its documentation.
A minor cleanup of the sherpa.estmethods code.
Allow a test to pass when given a relative path for the test data directory.
Update docs on how the docs are build in CI
Remove Meta class for header information that is used only rudimentary and
acts almost, but not quite, like a dict
Recent work on the ignore / notice logic in #1127 implicitly assumes that the
energy grid in PHA/ARF/RMF files is in increasing order, which does not hold
true for XMM/RGS data
For some data files, plots with a reversed grid (i.e. plots in wavelength) were
drawn with disjoint lines since #906.
Add a section on running the tests to the developer docs
Include the sherpa.utils.logging module in the documentation and update a test
to use the caplog feature of pytest (the code had been written to work with
python 2.7 version of unittest).
Updates to doc pages that list dangerously old numpy versions
Add a number of tests for corner cases of filtering and grouping of PHA data.
Improve the documentation of the filtering and grouping of PHA data.
Minor improvements to the documentation of the DataPHA class.
Change the internal storage handling of contour objects
Change the internal storage handling of image objects and improved testing
hc (as in "Planck constant * speed of light") is a physical constant that does
not depend on the DataPHA and thus it should not be defined as a class attribute.
Minor fixes and improvements to the documentation (both docstrings and RTD)
for the model- and parameter-related code.
Enable simulations with multiple ARF/RMF present (e.g. LETG/HRC).
Minor tweaks to the parameter tests.
The use of plot_model_component and get_model_component will now automatically
add the response for PHA data sets, if there is one. Model expressions which
contain a response will not be changed. Fixes #1020.
update to support environment changes due to conda defaults updates for
gcc 9.3.0
updates to switch python support to python 3.7-3.9
Improve datastack display (via show_stack and in the notebook) for datastacks.
Very minor improvement to the XSPEC test suite.
Very minor code reorganization for the regrid code.
updates gitlab conda recipe to python 1.19+ and adds run time pin
updates to fold in changes to CIAO region lib for CIAO 4.14. Specifically,
fixes for SM-89: bug in pie shape (extent/inside logic), SL-243: region -
deprecate obsoleted Warning.
Remove the SherpaTestCase class from sherpa.utils.testing and switch the tests
that used it to use pytest functionality where appropriate.
The get_xerr method of Data1DInt would fail if all bins had been ignored and
it now returns the empty list.
Ensure that the low and high limits for notice and ignore calls for DataPHA
objects are sensible (hi >= lo and that for energy or wavelength filters they
are >= 0). When filtering DataPHA objects in channel units the lo and hi
arguments must be integers otherwise an error is raised.
Energy and wavelength filters (with notice and ignore) for PHA data are now
treated as lo <= x < hi in all cases, and channel filters are lo <= x <= hi
(where the channel values must be integers). For Data1DInt cases the notice
and ignore filters are treated as lo <= x < hi. This changes address a number
of corner cases.
This can result in slightly different fit results because the data used in the
fit can be changed and hence changing the number of degrees of freedom (by one
or two depending on whether the first and last bins have been removed).
This change updates the version of the fftw module in sherpa's external
dependencies (extern) directory to utilize version 3.3.9 (www.fftw.org) instead
of the previous version 3.3.8
Filter expressions for Data1DInt and DataPHA cases now report the start and end
value of each set of grouped data rather than use the end points. For PHA data
sets users now see the same ranges displayed whether the data is grouped or not.
The actual filter remains the same even if the filter expressions has changed
slightly.
Support the new additive and multiplicative models in XSPEC 12.12.0 - XSgrbjet,
XSwdem, XSvwdem, XSvvwdem are additive and XSzxipab is multiplicative - and
note the additional abundance tables that are available when running with XSPEC
12.12.0 (lpgp and lpgs).
The AstroPy FITS reader is less tolerant of invalid FITS files in AstroPy 4.3
which trips one of our tests. The test has been updated with a FITS file that
doesn't trigger the error case. There is no functional change in this commit.
Fix three XSPEC models which had some internal confusion over parameter names:
XSzkerrbb should have uses fcol but had instead used hd, and XSzashift and
XSzmshift used Redshift instead of Velocity. The old names have been kept as
aliases.
Updates submodule to pick up the submodule README
Clean up of a test directory. There is no functional change in this commit.
Avoid compiler warnings from clang, fixing #1232 and several other warnings.
Initial support of sherpa for ARM build
In certain circumstances (such as plot_pvalue with an explicit 2D PSF
convolution operator read from a file) the code would fail with the error
unable to parse region string: None (issue #1214). This PR fixes this problem
and several minor related issues.
Update the image filter code so that the region description matches what the
filter actually is: rather than use & to combine regions we now use | as it
is a logical or rather than logical and. This is mainly going to affect users
who: have used the output of get_filter() as input to calc_data_sum2d, have
very-complex spatial filters, or are restoring Sherpa sessions created by save
or save_all.
Fix a problem with the output of the save_all command if an image had been
loaded but no region filter applied (fix #437): the output file would contain
a notice2d_id call with a filter of "" which would fail if you tried to load
the file back in. It now no-longer creates this line.
Update the documentation for set_stat to note you can send in a stats object
plot_model(), if done before calling notice or ignore, could lead to strange
filters (e.g. filter not getting applied). This has been fixed. Fixes #1024
Avoid warnings from NumPy 1.20 about using the deprecated np.int symbol.
Change the default X-Spec chatter level from 0 to 10. This may result in screen
output the first time an XSPEC model is evaluated (e.g. from a plot_model or
fit call). There have been a number of cases where APEC models appeared to be
creating no output because of missing data files (due to old ~/.xspec/Xspec.init
settings) which would have been more obvious with this change. To get back to
the previous behavior users can call set_xschatter(0).
XSPEC model parameters now use the same limits for soft and hard (the hard
limit from the model.dat file). This is handled by the XSBaseParameter class
which is used for XSPEC table models. The XSParameter class, which extends this
and is used by most other parameters, allows the user to change the hard limits
of the model. This follows XSPEC, and allows using some models which are
documented as supporting a value outside the normal parameter range (normally
by setting the value to a negative value). Note that this is potentially
dangerous (it could crash the program) so should be used carefully. It is
strongly suggested that any parameter set to a value outside of the original
limits is also frozen.
Add support for parsing XSPEC model.dat files (useful for updating the XSPEC
module or if you want to use XSPEC local models).
Update the XSPEC models to match the latest model.dat file (from HEASOFT 6.29 /
XSPEC 12.12.0). This primarily changes the hard limits of the models, but there
are some changes to the default parameter values as well, as well as some unit
changes, including adding and removing units.
It was possible to cause Sherpa to crash by passing an invalid function as the
groupfunc argument to apply_filter and apply_group. The code now errors out
instead. There are improvements to the documentation (docstring and RTD) for
PHA filtering and grouping.
Multiply the number of working references on readthedocs, while at the same
time reduce verbosity in the docstrings.
Update the Python supported status in the package metadata (drop 3.5, add 3.8)
and bump the minimum version to 3.6.
This change modifies the crates backend ascii file handling to resolve issue
#1262 where a test added in PR #1253 causes the tests to fail. This change may
potential result in behavior changes for crates use on ascii data files (see
caveat section for details).
Updates the data I/O so that write_arrays behaves consistently whether writing
ascii or fits files via the pyfits backend.
rm the platform module since it is not needed for ARM support
Fix a link on the Read The Docs pages from PR #1260
Include the Zenodo details for the 4.13.1 release into the hard-coded list of
releases.
This PR adds the option on how the optimization minim will behave if the free
parameter is beyond the limit. The default parameter reflect is True (will
reflect by an equal amount about the limit). If reflect is set to False then
model function will return DBL_MAX (~1e308) and therefore it will not be
included in the simplex.
Use unique_ptr rather than auto_ptr in the pileup code to avoid C++ warnings.
Fix #505
Fix up some links in the RTD documentation for simulating PHA data.
Corrected the order of the points, x, and y values when displaying a CDFPlot
object, improve support for a list argument, and made minor additions to the
documentation in the sherpa.plot module.
Add a reference to the OSTI.GOV technical report in the comments in the C++
code used for testing the optimization functions.
Fix tests that started to fail once #1246 was merged.
Add a section to the ReadTheDocs site explaining how to "optimize a function"
as this is something we can do but only really with the low-level code.
Fix a minor warning when generating the Read The Docs pages.
Rename the first two parameters of the XSPEC XSkerrconv convolution model from
Index/Index1 to Index1/Index2. Fixes #1275
Add documentation for the reflect keyword added in #1285 for the neldermead
optimizer
Added additional details regarding sherpa's MCMC inplementation to the
documentation
Fix the examples in the get_filter documentation.
Updates the minimum supported version of python supported in setup.py to use
python 3.7
- Crates behavior change - PR #1273 fixes an issue with the crates backend where supplying a column filter "opt colnames=none" for an ascii file would utilize default column names 'col1..coln' instead of the column names specified in the file. Code which relied on the old behavior may now produce an IO error such as "IOErr: Required column 'col1' not found in [ <list of valid column names>]"
Files
sherpa/sherpa-4.14.0.zip
Files
(15.5 MB)
| Name | Size | Download all |
|---|---|---|
|
md5:73bd7aed3c7fe03845b5931057d19c12
|
15.5 MB | Preview Download |
Additional details
Related works
- Is supplement to
- https://github.com/sherpa/sherpa/tree/4.14.0 (URL)