Romain Hugonnet

And I also have a question: How do we expect the profiling to work for parallelized functions (whether with Multiprocessing or Dask)?

@marinebcht That makes sense, otherwise many steps can't be visualized properly because the labels become too small. :sweat_smile: Let's add Plotly as an optional dependency! We'll also move Matplotlib to...

And for the tests of optional dependencies, we can do what was discussed here (we should do it in a separate PR, but relatively soon to ensure it doesn't break...

Looking good! We're going to have to prioritize tests with optional dependencies soon, to ensure at 100% that all of this runs even without `psutil` installed.

OK! 🙂

Another one here during CPD registration: https://github.com/GlacioHack/xdem/blob/49af1dd57c429475781b23a440a8d249501b4daa/xdem/coreg/affine.py#L1244

Actually, there is a difference between `_preprocess_subsample` applied to VS or to ICP/LZD/CPD: The `return_coords` argument is used for ICP/LZD/CPD, and not for VS. This could explain the high-memory usage....

Conclusion: I really don't know! 😅 It would be great to solve this. Ideally, point cloud methods like ICP should actually use *less* memory than grid-based method (like NK/LZD) as,...

Turns out after investigation using the profiling module: The big memory usage is the `apply()` part, using either `griddata` or our custom apply. For both, we could chunk them and...

Thanks @marinebcht, this is great! On the 15% performance reduction for "all": Interesting, I would have expected minimal overhead. I think this will greatly depend on the array size you...