WIP: Introduce plate class

[jo-mueller]

This is a first draft of an implementation of how writing plates could look like in a class-based API world, built around the NgffMultiscales class proposed in #515

[jo-mueller]

@will-moore if you're interested, this would be roughly my idea on how a class-based API could look like for writing HCS data (see classes/hcs.py)

In essence, one would have to create a dictionary of NgffMultiscales objects from #515 like this beforehand:

plate_dict = {
  ("A", 1): [ngff_rowA_col1_fov1, ngff_rowA_col1_fov2, ...], 
  ("A", 2): [ngff_rowA_col2_fov1, ngff_rowA_col2_fov2, ...],
  ("B", 1): [ngff_rowB_col1_fov1, ngff_rowB_col1_fov2, ...]
}

ngff_plate = NgffHCSPlate(plate_dict)
ngff_plate.to_ome_zarr(...)

this would then internally populate the omzp metadata, and dump everything to disk.

[will-moore]

So, you have to have the whole plate in memory / dask-arrays before you write?
That makes sense from an API point of view, but I'm not sure how practicable that is if you're writing a lot of data.

When we export Plates with omero-cli-zarr, we write each Image to disk at a time. https://github.com/ome/omero-cli-zarr/blob/9f844b877d8e19e42a69d7ba43f7dc086680fbec/src/omero_zarr/raw_pixels.py#L290
so it would be nice if that was an option.

In your plate_dict both row and column names should be strings. It might be nicer to simply use a key like "A/1" ?

[jo-mueller]

So, you have to have the whole plate in memory / dask-arrays before you write?
That makes sense from an API point of view, but I'm not sure how practicable that is if you're writing a lot of data.

Kind of. Under the hood, everything is coerced to a dask array, so the memory footprint should be low. Or at least not larger than whatever a user has in memory in the first place.

When we export Plates with omero-cli-zarr, we write each Image to disk at a time.

The to_ome_zarr function does effectively that. It iterates over all NgffMultiscales instances and if compute is True (which it defaults to) then every image is written at a time. You could set it to False and then write everything concurrently, but I'm not sure what one would gain from it 🤔

[imagesc-bot]

This pull request has been mentioned on Image.sc Forum. There might be relevant details there:

https://forum.image.sc/t/how-to-convert-czi-zeiss-to-ome-zarr-in-the-context-of-high-content-screening/120809/6