Fix SinglePlane example

[jeskowagner]

The SinglePlaneAcquisition example is currently broken in two ways:

1. The chunk dimensions do not match the image dimensions (i.e. specify a z-channel when there is none).
2. No distributed client is passed.

Traceback of first issue

AssertionError                            Traceback (most recent call last)
Cell In[15], line 4
      1 plate = converter.create_zarr_plate(plate_acquisition)
      3 # Run conversion.
----> 4 converter.run(
      5     plate=plate,
      6     plate_acquisition=plate_acquisition,
      7     well_sub_group="0",
      8     chunks=(1, 512, 512),
      9     max_layer=2,
     10 )

File <...>/src/faim_ipa/hcs/converter.py:152, in ConvertToNGFFPlate.run(self, plate, plate_acquisition, wells, well_sub_group, chunks, max_layer, storage_options, build_acquisition_mask)
    129 """
    130 Convert a plate acquisition to an NGFF plate.
    131 
   (...)
    149     zarr.Group of the plate.
    150 """
    151 assert 2 <= len(chunks) <= 3, "Chunks must be 2D or 3D."
--> 152 assert len(chunks) == len(
    153     plate_acquisition.get_well_acquisitions()[0].get_tiles()[0].shape
    154 ), "Chunks must have the same number of dimensions as the tile shape."
    155 well_acquisitions = plate_acquisition.get_well_acquisitions(wells)
    157 for well_acquisition in well_acquisitions:

AssertionError: Chunks must have the same number of dimensions as the tile shape.

Fixing this line to read chunks=(512, 512) lets us progress to the second issue:

Traceback of second issue

AttributeError                            Traceback (most recent call last)
File <...>/src/faim_ipa/hcs/converter.py:164, in ConvertToNGFFPlate.run(self, plate, plate_acquisition, wells, well_sub_group, chunks, max_layer, storage_options, build_acquisition_mask)
    158 well_group = self._create_well_group(
    159     plate,
    160     well_acquisition,
    161     well_sub_group,
    162 )
    163 group = well_group[well_sub_group]
--> 164 self._write_stitched_image(
    165     group,
    166     chunks,
    167     plate_acquisition,
    168     storage_options,
    169     well_acquisition,
    170     build_acquisition_mask=build_acquisition_mask,
    171 )
    172 shapes, datasets = self._build_pyramid(
    173     group,
    174     chunks,
    175     max_layer,
    176     storage_options,
    177 )
    178 self._write_metadata(
    179     group, max_layer, shapes, datasets, plate_acquisition, well_acquisition
    180 )

File <...>/src/faim_ipa/hcs/converter.py:241, in ConvertToNGFFPlate._write_stitched_image(self, group, chunks, plate_acquisition, storage_options, well_acquisition, build_acquisition_mask)
    238 rechunked_da = binned_da.rechunk(self._out_chunks(binned_da.shape, chunks))
    239 options = self._get_storage_options(storage_options, rechunked_da.shape, chunks)
    240 wait(
--> 241     self._client.persist(
    242         da.to_zarr(
    243             arr=rechunked_da,
    244             url=group.store,
    245             compute=False,
    246             component=str(Path(group.path, "0")),
    247             storage_options=options,
    248             compressor=options.get(
    249                 "compressor", zarr.storage.default_compressor
    250             ),
    251             dimension_separator=group._store._dimension_separator,
    252         ),
    253     )
    254 )

AttributeError: 'NoneType' object has no attribute 'persist'

Which seems to be triggered by converter._client being None by default. Initalising the client like in this example fixes that issue, too.

This PR includes both required changes.

[tibuch]

Hi @jeskowagner,

Thanks a lot for fixing this example.

We haven't updated the examples in a long time and most likely none of them are working anymore 🙃

I am very happy to see that other people are using this as well 🥳

[imagejan]

Funny coincidence, I was looking at this example today as well 😄

I'd suggest to also remove the stitching_yx_chunk_size_factor=2, I'll follow up next week on this.

[jeskowagner]

Yeah it seems we happened to work on this at the same time! Addressed in this new PR.