4.0.0 - 2022-10-11

Added

• Anaconda environments. There are two anaconda environments included
  inside the source folder at the moment. anaconda-environment.yml and
  dev-environment.yml. These should provide a nice starting point for
  anyone wanting to work on or with PyPWA. Pull requests are welcomed
  if you think a package should be added to the base environment.
• Added PyTorch for GPU and Apple Metal support. Can be specified during
  install using pip install pypwa[torch]. Amplitude support is specified
  by setting the USE_TORCH flag to True.
• Added support for Python's Multithreading. You should only use this when
  computation is happening on separate nodes and/or your optimizer choice
  does not support passing it's values across an OS Pipe.
• Added support for Minuit's parameter array argument. Now amplitudes can
  be written to accept a single array containing all the array values.
• Debugging support for amplitudes is now explicit. You can set the DEBUG
  flag to True on your amplitude before simulation or fitting, and it'll run
  in the main process so traceback and errors will not be suppressed.
• Amplitudes can now know where they live. Amplitudes have a THREAD flag
  that is numbered from 0 to N-threads that will specify which thread the
  amplitude is running in. This is useful if you want to pair your
  processes/threads with external devices like GPUs or OpenMPI nodes.

Changed

• Data module will no longer bury the Cache object. The cache object will
  now reside in the same directory as the parsed data.
• Moves Emcee to an optional dependency so that PyPWA can function in a
  base anaconda environment. If Emcee is installed, or if emcee is
  specified during installation using pip install pypwa[emcee], the
  emcee functionality will be usable.
• iMinuit has changed their ABI entirely, so the iminuit function has
  been changed to adapt to the new ABI.
• Updated all dependencies around ReadTheDocs to avoid GitHub flagging
  the dependencies for exploits.

Removed

• Project manager. There were several bugs throughout the module, and as
  far as we are aware, no users using the module. If you're affected by
  this change, please open an issue in the issue tracker to let us know.
• Removed the command line Binning utility. The Jupyter-based and
  internal binning utilities remain unaffected. If this affects you,
  please open an issue.
• Removed appdirs as a dependency.
• Removed CuPy support, replaced by PyTorch.
• Removed PyYaml Configuration support.

Fixed

• The bin by range function was not sampling data correctly. The intended
  behaviour was for each bin to be sampled by N samples, and then those
  samples to be shuffled to add randomization. However, because the
  shuffling was improperly implemented, what would occur instead is a single
  random event would be dropped from the sample, and then returned. This
  no longer occurs, and the returned bins will now be the correct length,
  and will be correctly shuffled.