Not clear how to update a tile

[surya-narayanan]

Following instructions in #281, I want to update tiles with outputs of a new transform.

My code is as follows.

pipeline = Pipeline([
    CollapseRunsVectra(),    
    SegmentMIF(model='mesmer', nuclear_channel=0, cytoplasm_channel=2, image_resolution=0.5, 
              postprocess_kwargs_whole_cell=None, 
              postprocess_kwrags_nuclear=None),
    QuantifyMIF('nuclear_segmentation')   
    ])


for i, h5_path in enumerate(h5s):  
  #Create slide object
  h5 = SlideData(h5_path)
  tile_keys = list(h5["tiles"].keys())
  tiles = [h5["tiles"][str(k)]["array"][:] for k in tile_keys]

  for tile in tiles:
    pipeline.apply(tile)
    h5.update(tile) #not sure if this is the right way to do things here

I am not able to follow the documentation and the code

pathml/pathml/core/tiles.py

Line 82 in acaef6c

def update(self, tile):

I am looking for an update function that takes two arguments, a key and a value, and updates the h5's tiles at the key, with the value. Now, it seems to only take a key object. I am not sure what it updates the tile with.

This would require me to make a new Tile() object before updating the h5, is that right?

If I am wrong, please tell me how to update the h5 data with the new tile.

[ryanccarelli]

The update function takes a Tile object https://pathml.readthedocs.io/en/latest/api_core_reference.html?highlight=Tiles#pathml.core.Tile

Internally tile.coords is the key.

You can simplify by iterating over SlideData.tiles:

pipeline = Pipeline([
    CollapseRunsVectra(),    
    SegmentMIF(model='mesmer', nuclear_channel=0, cytoplasm_channel=2, image_resolution=0.5, 
              postprocess_kwargs_whole_cell=None, 
              postprocess_kwargs_nuclear=None),
    QuantifyMIF('nuclear_segmentation')   
    ])

for h5_path in h5s:  
  #Create slide object
  h5 = SlideData(h5_path)
  for tile in h5.tiles:
    pipeline.apply(tile)
    h5.tiles.update(tile)

I agree that documentation can be improved. We should say tile (pathml.core.tile.Tiles): Updated tile. The tile with matching tile.coords will be replaced

[jacob-rosenthal]

Thanks Surya!
Looking at the code you posted, I wonder if this line might be causing a problem:

tiles = [h5["tiles"][str(k)]["array"][:] for k in tile_keys]

It seems like this is trying to make a list holding all tiles - which could be a problem if the tiles don't all fit in memory. Perhaps you could try turning it into a generator expression (i.e. tiles = (h5["tiles"][str(k)]["array"][:] for k in tile_keys) ) or using the for loop like Ryan suggested, both of which should retrieve tiles lazily one at a time instead of all at once

[surya-narayanan]

Yes. I used Ryan's suggestion to get it working. Thanks.