✔️ Linting Passed

All linting checks passed. Your pull request is in excellent shape! ☀️

_{Generated for commit: 32e9f17. Link to the linter CI: here}

PyPy tests go until 98%, almost there!

Is the plan to merge this even before we get to 100%? From a quick look at the diff it looks uncontroversial. I think technically it changes the API of the loss functions (they now return None) but I think they are internal so it isn't something we need to worry about.

@lorentzenchr do you want to have a look and think about this (I think you wrote a lot of this code?)?

We can remove those return statements, I guess the only use is testing and the Python loss.py submodule.
I had them there because I don’t like Fortran style coding and it made using those functions more Pythonic.

The docstrings need to be modified, too.

I assume the fortran style is needed to make sure we re-use the same array/memory each time and results in a significant performance improvement. Is that the right assumption? I don't know enough about the history of this code and the discussions around it, so if this has been discussed before: I don't want to re-start an old discussion, just check that someone thought about this topic. Because I agree having a more pythonic API would be nicer.

I assume the fortran style is needed to make sure we re-use the same array/memory each time and results in a significant performance improvement.

We want to achieve ufunc like behaviour, e.g. np.add(a, b, out=c) and provide array c instead of creating it each time (c can be element-wise loss or a gradient). This saves memory and the time to allocate that memory each time. It also suits perfectly in our fitting alogs, where we usually only need the current gradient (and sometimes the previous one, but not more).

This saves memory and the time to allocate that memory each time

I also like the fact that we allocate the memory via NumPy in Python usually and just interact with it. We don't have to free the buffer ourselves.

Merging, thanks! Let's see if the PyPy build manages to complete from time to time with this fix 🤞 !