Documentation inconistency

[blaginin]

• TIA Toolbox version: 1.3.1, dev
• Python version: 3.10.8 (conda)
• Operating System: MacOS 13.1 (22C65)

Description

Trying to learn the library and discovered some cases where docs are inconsistent with the current behaviour.

What I Did

• Tried to create AnnotationStore (as it was stated here, here, here, here, here, and here).

>>>  store = AnnotationStore()
TypeError: Can't instantiate abstract class AnnotationStore with abstract methods __getitem__, __len__, __setitem__, commit, dump, dumps, open

AnnotationStore has become abstract, and it would be helpful to update the docs accordingly. Examples can either state that users must use derived classes or explicitly show methods usage on SQLiteStore or DictionaryStore.

• Tried to add an object to my AnnotationStore (as it was stated here, here, here, here, here, and here).

>>> store = DictionaryStore()
>>> store.add(
...     Annotation(
...         geometry=Polygon.from_bounds(0, 0, 1, 1),
...         properties={"class": 42},
...     ),
...     key="foo",
... )
AttributeError: 'DictionaryStore' object has no attribute 'add'

It looks like .add → .append?

• Tried to query my AnnotationStore (as it was stated here and here).

>>> store = DictionaryStore()
>>> store.append(
...     Annotation(
...         geometry=Polygon.from_bounds(0, 0, 1, 1),
...         properties={"class": 42},
...     ),
...     key="foo",
... )
>>> store.bquery(where="props['class'] == 42") 
AttributeError: 'NoneType' object has no attribute 'bounds'

The method works successfully with SQLiteStore but fails on DictionaryStore. Maybe it would be helpful to add a test for that behaviour.

• Tried to predict classes of batches (as it was stated here)

>>> wsi_data = wsi.slide_thumbnail(resolution=5, units="power")
>>> assert type(wsi_data) == np.ndarray

>>> predictor = PatchPredictor(pretrained_model="resnet18-kather100k", batch_size=32)
>>> tile_output = predictor.predict(
...    imgs=[wsi_data],
...    mode="tile",
...    merge_predictions=True,
...    patch_input_shape=[224, 224],
...    stride_shape=[224, 224],
...    resolution=1,
...    units="baseline",
...    return_probabilities=True,
...    on_gpu=False
... )
TypeError: expected str, bytes or os.PathLike object, not ndarray

The method works with filenames but not with ndarray objects. Thus, users have to save their intermediate images to call the methods, which seems like an extra step.

[shaneahmed]

@blaginin Please create a PR to suggest edits to the documentation. You can also look at #540 if you have any suggestions.

[shaneahmed]

Hi @blaginin, Do you have any comments on #565 ?

[blaginin]

Hi @blaginin, Do you have any comments on #565 ?

@shaneahmed, all fixed except the last point. I can change it in a separate PR.

[John-P]

👍 Yep, I think the last point should be address in a separate PR (and probably a separate issue to keep track and so this one can be closed).