Graph array multiscale

[yfukai]

This pull request introduces support for arbitrary strides in the GraphArrayView class, allowing for efficient downsampling and subsetting of graph-based array data. I'm planning to benchmark this approach and mark this ready for review if I find this approach efficient.

[JoOkuma]

That will be awesome, @yfukai.
Feel free to add the benchmark to the benchmark folder.

It's very rudimentary, and we don't automatically check in the CI if the performance gets worse, but it's still a way to keep track of it.

[JoOkuma]

@yfukai, I don't know if it's faster or slower, but would it make sense to use this

        def mask_indices(
                self,
                offset: NDArray[np.integer] | int = 0,
                strides: NDArray[np.integer] | int = 1,  # new parameter
        ) -> tuple[NDArray[np.integer], ...]:

With the previous implementation?
b8617fc

I remember you updated this because the binary mask approach was much faster.

[JoOkuma]

and use this only if strides != 1

[yfukai]

Thanks for your suggestion! Maybe we should benchmark different approaches.

[yfukai]

Hi thanks @JoOkuma! Would it be fine if I rewrite or add benchmarks using asv?

[JoOkuma]

Hi thanks @JoOkuma! Would it be fine if I rewrite or add benchmarks using asv?

@yfukai, that would be awesome! Can you do it in a separate PR?