Packaging scientific functionality encapsulated in Mantid?

[SimonHeybrock]

Idea

• Try on a simple example and then evaluate

Requirements

• Usable without Mantid
• Usable without scipp
• Usable by Mantid and scipp
• Installable via pip

Questions

Now

• Interface/API? Numpy arrays?
• What is a good candidate algorithm?
  • mantid.CylinderAbsorption?
  • mantid.PaalmanPings
  • Above probably needs MantidGeometry, are there alternatives?
    • Yes, nominated https://docs.mantidproject.org/nightly/algorithms/CylinderPaalmanPingsCorrection-v2.html as good candidate.
  • Fitting or fit functions
    • Ikeda-Carpenter
  • Single crystal, peak integration?

Later

• How to define boundaries? Which components should be part of the same package?
• Which dependencies (on other Python software) should be allowed?

Outcomes

• Cost of setting up CI, packing, ...
• Cost of creating wrappers in scippneutron
• Cost of creating wrappers in Mantid
• Performance/memory overhead caused wrapping

[SimonHeybrock]

• Common scientific base. Document what we know and what we do not know, such as magic constants.
  • Implemented as generic function -> unit tests as well as "manual" verification, e.g., showing that output matches measured tube efficiency or absorption.
  • Could be used directly by libraries that implement array-function protocol.
  • Define schema of what needs to be documented (should be machine-readable, or accessible in the package)?:
    • What assumptions were made?
    • How was it verified?
    • Are there unknown magic constants?
    • Precise version (and hash?) for FAIR / reproducible results, years after original data reduction
    • Ability to flag versions as broken/incorrect?
    • ...
• Do we need specification similar to Python array API standard, but science-focused, e.g., with units and labelled dimensions?
• Are event-data copies too inefficient to be worthwhile?
• Is implementing the NumPy __array_function__ a solution?
• Documentation duplication is an as-big, or even bigger, issue than code duplication.