History

For a brief overview of the history and state-of-the-art in spectral timing, and for more information about the design and capabilities of Stingray, please refer to Huppenkothen et al. (2019).

Stingray originated during the 2016 workshop The X-ray Spectral-Timing Revolution: a group of X-ray astronomers and developers decided to agree on a common platform to develop a new software package. At that time, there were a number of official software packages for X-ray spectral fitting (XSPEC, ISIS, Sherpa, …), but such a widely used and standard software package did not exist for X-ray timing, that was mostly the domain of custom, proprietary software. Our goals were to merge existing efforts towards a timing package in Python, following the best guidelines for modern open-source programming, thereby providing the basis for developing spectral-timing analysis tools. We needed to provide an easily accessible scripting interface, a GUI, and an API for experienced coders. Stingray’s ultimate goal is to provide the community with a package that eases the learning curve for advanced spectral-timing techniques, with a correct statistical framework.

Further spectral-timing functionality, in particularly command line scripts based on the API defined within Stingray, is available in the package HENDRICS. A graphical user interface is under development as part of the project DAVE.

Previous projects merged to Stingray

  • Daniela Huppenkothen’s original Stingray

  • Matteo Bachetti’s MaLTPyNT

  • Abigail Stevens’ RXTE power spectra code and phase-resolved spectroscopy code

  • Simone Migliari’s and Paul Balm’s X-ray data exploration GUI commissioned by ESA

Changelog

v2.2 (2024-10-22)

New Features

  • Add a compute_rms function to LombScarglePowerspectrum (#828)

  • Introduced FITSReader class for lazy-loading of event lists (#834)

  • implementation of 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

  • Eliminated runtime dependency on setuptools (#852)

  • Moved configuration to pyproject.toml as recommended by PEP 621 (#842)

  • Added pre-commit hooks in pre-commit-config.yaml (#847)

  • Improved main page of the documentation (#748)

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 with np.asanyarray everywhere, to avoid copying memory maps into memory if possible (#824)

Internal Changes

  • Dead time model fixes: more stable computations, better plotting of check_A and check_B (#800)

  • Bumped jinja version to 3.1.4 (#825)

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 corresponding LombScargleCrossspectrum and LombScarglePowerspectrum (#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. We now check for this situation and give 0 with an uncertainty as a result. (#736)

  • Eliminates deprecated call to enable_deprecations_as_warnings, 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, 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 from compute_rms. (#792)

Internal Changes

  • Speedup creation of events in EventList.from_lc (#757)

  • Separate slow tests from quick ones (#758)

  • Use Readthedocs for documentation building (#769)

  • More informative GTI messages (#787)

  • Eliminated the usage of astropy logging (#799)

v1.1.2 (2023-05-25)

New Features

  • Phase Dispersion Minimization as a method to search for periodic signals in data is now implemented in the stingray.pulse submodule. To use it, you can use the phase_dispersion_search function in stingray.pulse.search. The accompanying statistical tests are located in the stingray.stats module, under phase_dispersion_probability, phase_dispersion_logprobability and phase_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 to pyproject.toml and ignoring PEP-8 non-compliant E203, W503 in flake8. (#715)

  • Added a scrollbar to sidebarwrapper (#718)

  • Simplify numba mocking code, and possibly improve code coverage estimate (#731)

v1.1.1 (2022-10-10)

Bug fixes

  • Fixed white_noise_offset in compute_rms to 2.0, as it should be

  • Fixed a bug that produced a crash when calculating the rms in spectra corrected through the FAD technique

  • Fixed a bug that eliminated the imaginary part from cross spectra corrected with the FAD

  • Fixed a bug that considered contiguous GTIs as non-continuous (due to very small differences between stop and start of the next GTI) by allowing a small tolerance

Full list of changes

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 for light curves.

Improvements

  • A less confusing default value of segment_size in Z searches

Full list of changes

v1.0 (2022-03-29)

TL,DR: these things will break your code with v1.0:

  • Python version < 3.8

  • The gtis keyword in pulse/pulsar.py (it is now gti, 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 modlation 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 and LightCurve objects into Astropy TimeSeries

  • New infrastructure for converting most Stingray classes into Astropy Table objects, Xarray and Pandas data frames.

  • 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 in DynamicalPowerSpectrum

  • New stingray.fourier module containing the basic timing products, usable on numpy arrays, and centralizes fft import

  • New methods in Crossspectrum and Powerspectrum 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 deprecated 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 and AveragedPowerspectrum 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 keyword gti instead of gtis (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, and from_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

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)

  • Fixed a bug in which all simulator errors were incorrectly non-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 list of changes

v1.0beta was released on 2022-02-25.

v0.3 (2021-05-31)

  • 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

Full list of changes

v0.2 (2020-06-17)

  • Added Citation info

  • Fixed various normalization bugs in Powerspectrum

  • Speedup of lightcurve creation and handling

  • Made code compatible with Python 3.6, and dropped support to Python 2.7

  • Test speedups

  • Dead time models and Fourier Amplitude Difference correction

  • Roundtrip of LightCurve to lightkurve objects

  • Fourier-domain accelerated search for pulsars

  • Adapt package to APE-17

  • Periodograms now also accept event lists (instead of just light curves)

  • Allow transparent MJDREF change in event lists and light curves

Full list of changes

v0.1.3 (2019-06-11)

  • Bug fixes

v0.1.2

  • Bug fixes

v0.1.1

  • Bug fixes

v0.1 (2019-05-29)

  • Initial release.

Presentations

Members of the Stingray team have given a number of presentations which introduce Stingray. These include: