pyFAI 2023.1

First stable version of pyFAI: v2023.1.

Sources and binary wheels can be found at: https://pypi.org/project/pyFAI/
and at: https://github.com/silx-kit/pyFAI/releases/tag/v2023.1

One of a few ways to install this release with pip:

pip install pyFAI==2023.1.0

Release notes and important changes since 0.21:

• Developer and packager tools:
  • Switch build system from numpy.distutils to meson-python
  • Keep the former setup.py for compatibility reasons: it will be removed in a future release
  • Drop Python 3.6 (default parameters in namedtuple feature used)
  • Require silx 1.1 (for OpenCL), scipy and matplotlib
• GUI side:
  • several minor improvements in pyFAI-calib2
  • Fixed calibration in jupyter-lab
• Core improvements:
  • Refactoring of the Geometry class
  • Geometry pseudo-inversion optimization
  • Improved support from Medipix-based Lambda-detectors
  • New detectors from Dectris (Pilatus 900k and Eiger 250k)
  • Support Nexus format in output: NXmonpd and NXcansas
  • Single-threaded CSC sparse matrix multiplication engine
• Improved uncertainty propagation:
  • Refactor error model management (uses enum)
  • Hybrid error model (azimuthal for sigma-clipping but reports Poissonian noise)
  • Export peakfinder data to the CXI format (used by CrystFEL)
• Improvement in the doc:
  • Update installation instructions
  • Multi-threaded integration tutorial
  • GPU implementation tutorial
• Facts and figures:
  • 400+ commits, 100 PR
    +with the contribution of:
    Clemens Prescher,
    Elena Pascal,
    Jérôme Kieffer,
    Malte Storm,
    Marco Cammarata,
    Michael Hudson-Doyle,
    Picca Frédéric-Emmanuel,
    Rodrigo Telles,
    Thomas A Caswell,
    Tommaso Vinci,
    Valentin Valls,
    Wout de Nolf.

PyFAI v0.21.3

• Support hexagonal pixels in splitbbox and nosplit mode (collaboration with EuXFEL)
• Support extra dtype in OpenCL (contribution from Desy)
• Fix version of setuptools (enables to build pyFAI -> python 3.12)
• Support some of the Lambda detectors from XSpectrum (collaboration with Xmas)
• Restore back compatibility in histoBBox1d and histoBBox2d (contribution from NSLS-II)
• Fix bug in OpenCL distortion correction (collaboration with Soleil)
• Several other bugs fixed

PyFAI version 0.21.2

This is a doc-fix minor release:
Documentation for sparsify-Bragg and peakfinder CLI tools were corrected.

PyFAI version 0.21.1

This is a bug-fix release over 0.21:

• fix several cabling issue inside AzimuthalIntegrator, preventing unattended engine-reset
• dummy value was not working properly in Worker
• Remove deprecation warning about scipy
• Using silx for median-filter which is faster
• skip 3 tests which are known to be fragile on 32-bit platforms (spotted by debian on i386 and armel)

PyFAI version 0.21.0

• Sigma-clipping allows separation of Bragg/amorphous signal:
  • Implementation in Python, Cython and OpenCL with poissonian and azimuthal error-model
  • Sparsification, compression of single crystal data
  • Application to peak-picking and quality assessment of SSX data
  • Analysis of grid-scan to find single crystal
  • Single pass variance evaluation in azimuthal bin
  • Integration of the Jungfrau detector (ID29)
• 2D integration:
  • New integrators with error propagation by default
  • Full pixel splitting in addition to BBox and no splitting
• Refactor of all LUT and CSR to share the same code base, makes tests more robust.
• Calibration of experimental setup using Jupyter notebooks
  • User interaction with plots in matplotlib (thanks Phil Hans)
  • Factorize code between pyFAI-calib and jupyter calibration
  • Tutorial as notebook and video recording
• Parallax correction for thick detector (still experimental, thanks to Vadim)
• Improved tutorial on detector geometry calibration (based on Kabsch alignment)
• Better performances on HPC nodes by limiting simple OpenMP to fewer cores
• Many improvement in test, typos fixed, doc ...
• Deprecate all legacy integrators since the new version sees its matrix complete.
• Supports Python 3.6 ... 3.10 under Windows, MacOS and Linux.

PyFAI version 0.20.0

• One year of development, about 500 commits & 370 files modified.
• Generalization of the new generation 1D integrators (better error propagation)
• Sigma clipping and sparsification of single crystal data (OpenCL only)
• Fix issue introduced with the scipy 1.15 (constrained calibration broken)
• Improved distortion correction (also on GPU, ...)
• Major re-work of the documentation (thanks Thomas Kluyver and Loic Huder)
• Improve the calibration of Pilatus and Eiger detectors based on a grid of holes.
• New cylindrical detector from Rigaku
• Drop deprecated OpenCL integrator
• Support all Eiger2 detectors (thanks Clemens Weninger and Marie Ruat)
• CI: move to Gihub workflow and gitlab-runners (bob) for building (thanks Thomas Vincent).
• Build for debian 10 and 11 (also ubuntu 20.04), drop debian 9
• Remove Python2 related code
• Lower Numpy ABI dependency as much as possible (remains _distortion's C++ code)
• Drop tests for Python 3.5, checked on 3.9 as well.

PyFAI v0.19.0 31/01/2020

• Minor revison with only 150 commits, mainly bug-fixes
• Improvement on the GUI with many small bug-fixes
• Support newer h5py (mode mandatory, [()], ...)
• Build for debian 10 and 11 (also ubuntu 20.04)
• Drop tests for Python 2 and 3.4, checked on 3.8
• Improved compatibility with ImageD11
• Use hdf5plugin to provide hdf5 io-filters in apps
• Rework diffraction mapping tools to use a worker
• New generation azimuthal integrator using CSR algorithm
  implemented in Python, Cython and OpenCL.
• Sigma-clipping implemented in OpenCL
• Publication on new generation integration, the GUI for
  calibration and the goniometers accepted in J. Synch. Rad.
• A big thank to Florian from Germany, Bertrand from Xenocs,
  Alex from Soleil and Jon from ESRF for their contributions

PyFAI v0.18.0

• Last release with Valentin as he finishes his contract soon
• Almost 800 commits, 60 PR since the last release: this is a huge
  release !
• Major rework on all GUIs, mainly pyFAI-calib2 and pyFAI-integrate.
• Possibility to integrate image stacks (i.e. from HDF5), ...
• Rework the method to specify the algorithm, pixel splitting and
  implementation
  with sensible fall-backs. Also available via the different GUIs
• 3D visualization of detectors and experimental setup, useful for non
  flat detectors.
• integrate1d_ng is available with histogramming without pixel
  splitting in
  Python, Cython and OpenCL. Now, propagates the variance properly !
• IO sub-packages with associated refactoring for ponifile, json,
• Improved management of OpenMP: simplify the code for histograms.
• Improved geometry description and tutorial for writing exchange
  with other software (ImageD11, thanks to Carsten Detlefs).
• More reliable simple ellipse fitting with tests and doc.
• Better POCL integration (debugged on cuda, x87, Power9, ...)
• Rely on silx mechanics for the build, test, download, GUI,
  opencl ...
• Many new tutorials, now available on binder-hub: new calibrants,
  Pilatus calibration, ...
• Fix many issues reported in third-party software (Dioptas, ...)
• Drop support of debian8, Python 2.7 and Python 3.4 on all
  platforms. It is likely most functionalities still work but without
  guaranty.

PyFAI v0.17.1

Bug-fix release. Allows to build with numpy v1.16

PyFAI v0.17.0

• Only 200 commits in a couple of month, this is a small release
• Fix major bugs in pyFAI-calib2 (double validator, initial guess, ring position)
• Constrains have been added to the geometry fitting of pyFAI-calib2
• New pyFAI-integrate graphical application
• Much better support for user defined detector (HDF5)
• Start the rewrite of all integrators to allow proper error propagation (2D done)
• Factorize the preprocessing steps for many integrators
• Remove tons of code which has been deprecated for years in AzimuthalIntegrator
• Featuring contribution from Soleil and Berkeley
• Stop supporting Python2.7 on Windows
• All scripts are now using Python 3.x (x>=4)
• This is the last release supporting Python 2.7, 3.4 hence debian 8