Summary of speedup shapes layer project

[Czaki]

No description provided.

[jni]

@Czaki is this more or less done? Do you want me to have a pass at it?

[melonora]

Think overall it looks good, but I did improve the wording I think:) Also I added some information about the binary search trees as this is technical jargon that not everyone would understand. Please check if everything still aligns with your additional thoughts or a I made mistakes.

One thing I think you also have to decide I think is whether to write it not from a third perspective and your own perspective combined, but just either one of those (I would vote in hindsight after reviewing that your own perspective is probably best, also allows you to take ownership:)).

[jni]

I'll have another pass at the language, but one thing I'd like to see is a summary plot showing, for each improvement step, how much a particular benchmark (probably the real SpatialData dataset) improved.

I also prefer to put real numbers and code to claims like "np.unique is really slow" and "np array creation is really slow". A good illustration would be a section on normalize_vertices_and_edges, comparing my original NumPy-and-SciPy version (which I thought was pretty clever 😂) with Grzegorz's pure Python loop version, which was pretty much a straight port of the numba code yet still faster than all the SciPy acrobatics. Putting a section about that with numbers would be very compelling I think.

Finally, I definitely think we should change from third person to first person. It feels much more engaging to write from first person — people go through the journey with you. I'm happy to work on the language if you can work on the figures and benchmarks @Czaki!