Releases: StingraySoftware/stingray
Stingray v2.2
This release accompanies the review of Stingray from pyOpenSci. As such, it contains some internal changes that were driven by the review together with some new addition.
New Features
- Introduce FITSReader class for lazy-loading of event lists
(#834) - Add a compute_rms function to LombScarglePowerspectrum
(#828) - Implement the shift-and-add technique for QPOs and other
varying power spectral features
(#849)
Bug Fixes
- The
fold_events
function now checks if the keyword arguments
kwargs
are in the list of optional parameters. If
any unidentified keys are present, it raises a
ValueError. This fix ensures that the function only
accepts valid optional parameters and provides a clear error message
for unsupported keys.
(#837)
Internal Changes
Release 2.1
v2.1 (2024-05-29)
New Features
- Add function to calibrate event lists based on RMF file (#804)
- Speed up computation of pds for large arrays (#808)
- Add support for XTE science event data (#816)
- A friendlier API for the non-paralyzable dead time model model (#800)
Bug Fixes
- Fix issue when setting a property from a FITS file read (#814)
- Fix case when analyze_segments has an invalid segment (#822)
- Substitute
np.asarray
withnp.asanyarray
everywhere, to avoid copying memory maps into memory if possible (#824)
Internal Changes
Stingray Version 2.0
v2.0 (2024-03-13)
New Features
- Power colors à la Heil et al. 2015
(#780) - Calculate colors and intensities on a segment-by-segment basis in
event lists
(#781) - A function to randomize data in small bad time intervals
(#782) - The Lomb Scargle Fourier Transform (fast and slow versions) and the
correspondingLombScargleCrossspectrum
{.interpreted-text
role="class"} andLombScarglePowerspectrum
{.interpreted-text
role="class"}
(#737) - A JAX implementation of the Gaussian Process tool by Hubener et
al for QPO detection and
parameter analysis.
(#739) - Extend join operation for events to arbitrary array attributes, not
just pi and energy
(#742) - Allow the creation of empty light curves.
(#745) - Make StingrayTimeseries into a generalized light curve, with a less
strict naming but implementing much of the underlying computing
useful for Lightcurve as well.
(#754) - Our fast implementation of histograms is now safer (failing safely
to the equivalent numpy histogram functions), more consistent
(ranges moved to range, for consistency with numpy), and accept
complex weights as well!
(#764)
Bug Fixes
- When rms is low, the calculation in compute_rms often gave
[NaN]{.title-ref}. We now check for this situation and give 0 with
an uncertainty as a result.
(#736) - Eliminates deprecated call to
[enable_deprecations_as_warnings]{.title-ref}, and contextually,
changes the code to be much more robust in catching harmful
warnings.
(#738) - Changes Crossspectrum.plot() function to plot the actual real and
imaginary parts instead of their absolute values.
(#747) - Make commits marked as [docs only] skip all CI but the docs build
(#749) - Update tstart and tseg when using Lightcurve.truncate()
(#753) - Changed list comprehension to generator expression to reduce memory
usage.
(#756) - Fix a bug with segment sizes not exact multiples of dt when dealing
with light curves
(#760) - Fix a bug when light curve segments contain complex values
(#760) - Crossspectrum had "real" as default value. This meant that, for
example, lags could not be calculated. Now the default value is
"all", as it should be.
(#762) - Fix plotting of spectra, avoiding the plot of imaginary parts of
real numbers
(#763) - Various bugfixes in [gti.py]{.title-ref}, and a new function to
interpret the mix of multiple GTIs.
(#774) - Fixed subcs duplication by adding a check in the for loop that
copies the attributes from table's meta items.
(#776) - Various bug fixes in DynamicalPowerspectrum, on event loading and
time rebinning
(#779) - Fix issue with the Poisson noise calculation in lag spectra, that
produced NaN errors under some conditions
(#789) - Fix rms computation and error bars
(#792) - Fix issue with
Powerspectrum
of a single light curve
(#663) - Fix nphots estimate in accelsearch, that lead to an underestimation
of the power of candidates
(#807)
Breaking Changes
- Eliminate deprecated
format_
keyword from read and write methods.
(#729) - Remove legacy interface and obsolete large data machinery.
(#755) - Eliminate deprecated
white_noise_level
keyword fromcompute_rms
.
(#792)
Internal Changes
Stingray 2.0.0, release candidate 1
New Features
- Power colors à la Heil et al. 2015
(#780) - Calculate colors and intensities on a segment-by-segment basis in
event lists
(#781) - A function to randomize data in small bad time intervals
(#782) - The Lomb Scargle Fourier Transform (fast and slow versions) and the
correspondingLombScargleCrossspectrum
{.interpreted-text
role="class"} andLombScarglePowerspectrum
{.interpreted-text
role="class"}
(#737) - A JAX implementation of the Gaussian Process tool by Hubener et
al for QPO detection and
parameter analysis.
(#739) - Extend join operation for events to arbitrary array attributes, not
just pi and energy
(#742) - Allow the creation of empty light curves.
(#745) - Make StingrayTimeseries into a generalized light curve, with a less
strict naming but implementing much of the underlying computing
useful for Lightcurve as well.
(#754) - Our fast implementation of histograms is now safer (failing safely
to the equivalent numpy histogram functions), more consistent
(ranges moved to range, for consistency with numpy), and accept
complex weights as well!
(#764)
Bug Fixes
- When rms is low, the calculation in compute_rms often gave
[NaN]{.title-ref}. We now check for this situation and give 0 with
an uncertainty as a result.
(#736) - Eliminates deprecated call to
[enable_deprecations_as_warnings]{.title-ref}, and contextually,
changes the code to be much more robust in catching harmful
warnings.
(#738) - Changes Crossspectrum.plot() function to plot the actual real and
imaginary parts instead of their absolute values.
(#747) - Make commits marked as [docs only] skip all CI but the docs build
(#749) - Update tstart and tseg when using Lightcurve.truncate()
(#753) - Changed list comprehension to generator expression to reduce memory
usage.
(#756) - Fix a bug with segment sizes not exact multiples of dt when dealing
with light curves
(#760) - Fix a bug when light curve segments contain complex values
(#760) - Crossspectrum had "real" as default value. This meant that, for
example, lags could not be calculated. Now the default value is
"all", as it should be.
(#762) - Fix plotting of spectra, avoiding the plot of imaginary parts of
real numbers
(#763) - Various bugfixes in [gti.py]{.title-ref}, and a new function to
interpret the mix of multiple GTIs.
(#774) - Fixed subcs duplication by adding a check in the for loop that
copies the attributes from table's meta items.
(#776) - Various bug fixes in DynamicalPowerspectrum, on event loading and
time rebinning
(#779) - Fix issue with the Poisson noise calculation in lag spectra, that
produced NaN errors under some conditions
(#789) - Fix rms computation and error bars
(#792)
Breaking Changes
- Eliminate deprecated
format_
keyword from read and write methods.
(#729) - Remove legacy interface and obsolete large data machinery.
(#755) - Eliminate deprecated
white_noise_level
keyword fromcompute_rms
.
(#792)
Internal Changes
Version 1.1.2
New Features
- Phase Dispersion Minimization as a method to search for periodic
signals in data is now implemented in thestingray.pulse
submodule. To use it, you can use thephase_dispersion_search
function instingray.pulse.search
. The accompanying statistical
tests are located in thestingray.stats
module, under
phase_dispersion_probability
,phase_dispersion_logprobability
andphase_dispersion_detection_level
.
(#716) - Add is_sorted function, to test if an array is sorted.
(#723) - Check if invalid data are inside GTIs, and warn or raise exception
accordingly
(#730)
Bug Fixes
- The method apply_gtis of the class Lightcurve is applied to all the
attributes of the class Lightcurve. This works for both inplace=True
and inplace=False
(#712) - Avoid allocation of an unneeded square matrix to improve memory
management in_als
(fix Issue 724)
(#725) - Fix Issue #726 -- Loading events without fmt keyword crashes
(#727)
Documentation
- Reordered information about contributions with new
black
and
towncrier
procedures
(#721)
Internal Changes
- Using towncrier to generate
the changelogs.
(#697) - Added stingray's logo in the documentation's favicon and top bar.
(#707) - Improved contributing workflow by appending
black
codestyle
configuration topyproject.toml
and ignoring PEP-8 non-compliant
E203
,W503
inflake8
.
(#715) - Added a scrollbar to sidebarwrapper
(#718) - Simplify numba mocking code, and possibly improve code coverage
estimate
(#731)
v1.1
v1.1 (2022-10-02)
Bug fixes
- IMPORTANT: Fixed sign of time lags, which were calculated using the interest band as the reference.
- Fixed an issue when the fractional exposure in FITS light curves is slightly >1 (as sometimes happens in NICER data)
New
- Implemented the
bexvar
variability estimation method (Bayesian excess variance) for light curves.
Improvements
- A less confusing default value of
segment_size
inZ
searches
Version 1.0
TL,DR:
These things will break your code with v1.0:
- Python version < 3.8
- The
gtis
keyword inpulse/pulsar.py
(it is nowgti
, without the 's')
New
- Dropped support to Python < 3.8
- Multi-taper periodogram, including a Lomb-Scargle implementation for non-uniformly sampled data
- Create Count rate spectrum when calculating spectral-timing products
- Make modulation upper limit in (
Averaged
)Powerspectrum
work with any normalization (internally converts to Leahy for the calculation) - Implement Gardner-Done normalization (1 for perfect correlation, -1 for perfect anticorrelation) for
Auto
/Crosscorrelation
- New infrastructure for converting
EventList
andLightCurve
objects into AstropyTimeSeries
- New infrastructure for converting most Stingray classes into Astropy
Table
objects - Save and load of most Stingray classes to/from many different file formats (pickle, ECSV, HDF5, FITS, and all formats compatible with Astropy Table)
- Accept input
EventList
inDynamicalPowerSpectrum
- New
stingray.fourier
module containing the basic timing products, usable on numpy arrays, and centralizes fft import - New methods in
Crossspectrum
andPowerspectrum
to load data from specific inputs:from_events
,from_lightcurve
,from_time_array
,from_lc_list
(from_time_array
was also tested using memory-mapped event lists as inputs: useful in very large datasets) - New and improved spectral timing methods:
ComplexCovarianceSpectrum
,CovarianceSpectrum
,LagSpectrum
,RmsSpectrum
- Some depreciated features are now removed
PSDLogLikelihood
now also works with a log-rebinned PDS
Improvements
- Performance on large data sets is VASTLY improved
- Lots of performance improvements in the
AveragedCrossspectrum
andAveragedPowerspectrum
classes - Standardized use of new fast psd/cs algorithm, with legacy still available as an alternative option to specify
- Reading calibrated photon energy from event files by default
- In
pulse/pulsar.py
, methods use the keywordgti
instead ofgtis
(for consistency with the rest of Stingray) - Added more documentation about using GTIs for a
Powerspectrum
orCrossspectrum
object if you already passed a GTI to the inputLightcurve
object - Moved
CovarianceSpectrum
toVarEnergySpectrum
and reuse part of the machinery - Improved error bars on cross-spectral and spectral timing methods
- Measure absolute rms in
RmsEnergySpectrum
- Friendlier pyfftw warnings
- PDS/CrossSp initially store the unnormalized power, and convert it on the fly when requested, to any normalization
- Adjust codecov settings with new yaml file
Bug fixes
- Fixed error bars and error distributions for sliced (iterated) light curves and power spectra
- Fixed a bug in how the start time when applying GTIs (now using the minimum value of the GTI array, instead of half a time bin below the minimum value)
- All simulator errors are zero
- Fixed coherence uncertainty
- Documented a Windows-specific issue when large count rate light curves are defined as integer arrays (Windows users should use float or specify int-64)
- If the variance of the lightcurve is zero, the code will fail to implement Leahy normalization
- The value of the
PLEPHEM
header keyword is forced to be a string, in the rare cases that it's a number - and more!
Full Changelog: v0.3...v1.0
Version 1.0 beta
What's Changed
TL,DR: these things will break your code with v1.0beta:
- Python version < 3.8
- The
gtis
keyword inpulse/pulsar.py
(it is nowgti
, without the 's')
New
- Dropped support to Python < 3.8
- Multi-taper periodogram, including a Lomb-Scargle implementation for non-uniformly sampled data
- Create Count rate spectrum when calculating spectral-timing products
- Make modulation upper limit in
(Averaged)Powerspectrum
work with any normalization (internally converts to Leahy for the calculation) - Implement Gardner-Done normalization (1 for perfect correlation, -1 for perfect anticorrelation) for
Auto/Crosscorrelation
- New infrastructure for converting
EventList
andLightCurve
objects into AstropyTimeSeries
- New infrastructure for converting most
Stingray
classes into AstropyTable
objects - Save and load of most
Stingray
classes to/from many different file formats (pickle
,ECSV
,HDF5
,FITS
, and all formats compatible with Astropy Table) - Accept input
EventList
inDynamicalPowerSpectrum
- New
stingray.fourier
module containing the basic timing products, usable onnumpy
arrays, and centralizes fft import - New methods in
Crossspectrum
andPowerspectrum
to load data from specific inputs:from_events
,from_lightcurve
,from_time_array
,from_lc_list
(from_time_array
was also tested using memory-mapped event lists as inputs: useful in very large datasets) - New and improved spectral timing methods:
ComplexCovarianceSpectrum
,CovarianceSpectrum
,LagSpectrum
,RmsSpectrum
- Some depreciated features are now removed
PSDLogLikelihood
now also works with a log-rebinned PDS
Improvements
- Performance on large data sets is VASTLY improved
- Lots of performance improvements in the
AveragedCrossspectrum
andAveragedPowerspectrum
classes - Standardized use of new fast psd/cs algorithm, with
legacy
still available as an alternative option to specify - Reading calibrated photon energy from event files by default
- In
pulse/pulsar.py
, methods use the keywordgti
instead ofgtis
(for consistency with the rest of Stingray) - Moved
CovarianceSpectrum` to
VarEnergySpectrum`` and reuse part of the machinery - Improved error bars on cross-spectral and spectral timing methods
- Measure absolute rms in
RmsEnergySpectrum
- Friendlier
pyfftw
warnings - Streamline PDS/CrossSp production, adding
from_events
,from_lc
,from_lc_iterable
, andfrom_time_array
(to input a numpy array) methods - PDS/CrossSp initially store the unnormalized power, and convert it on the fly when requested, to any normalization
- Adjust codecov settings with new yaml file
Bug fixes
- Fixed error bars and
err_dist
for sliced (iterated) light curves and power spectra - Fixed a bug in how the start time when applying GTIs (now using the minimum value of the GTI array, instead of half a time bin below the minimum value)
- All simulator errors are zero
- Fixed coherence uncertainty
- Documented a Windows-specific issue when large count rate light curves are defined as integer arrays (Windows users should use
float
or specifyint-64
) - If the variance of the lightcurve is zero, the code will fail to implement Leahy normalization
- The value of the
PLEPHEM
header keyword is forced to be a string, in the rare cases that it's a number - and more!
Full Changelog: v0.3...v1.0-beta
Version 0.3
New features and changes:
- Lots of performance improvements
- Faster simulations
- Averaged Power spectra and Cross spectra now handle Gaussian light curves correctly
- Fixes in rebin functions
- New statistical functions for signal detection in power spectra and pulsar search periodograms
- Much improved FTOOL-compatible mission support
- New implementation of the FFTFIT method to calculate pulsar times of arrival
- H-test for pulsar searches
- Z^2_n search adapted to binned and normally distribute pulse profiles
- Large data processing (e.g. from NICER) allowed
- Rebinning function now accepts unevenly sampled data
- New saving and loading from/to Astropy Tables and Timeseries
- Improved I/O to ascii, hdf5 and other formats
- Rehaul of documentation, now on stingray.science
Version 0.2
Performance improvements:
-
Much faster loading of light curves
-
Faster testing
-
Faster pulsar searches
New functionalities:
-
Fourier-domain accelerated pulsar searches
-
Fourier Amplitude Difference correction for power density spectra affected by dead time
Infrastructure changes:
- Adaptation of package to APE17
Major bug fixes:
-
Rms Normalization of PDSs was wrong under certain conditions
-
A test for numpy complex types done wrong produced lag measurements identical to 0
Plus many, many smaller bug fixes and performance tweaks. Enjoy!