Crashes when creating TF record from joint dataset with some empty engraving categories

[v-dvorak]

The error

Traceback (most recent call last):
  File "create_tf_record_from_joint_dataset.py", line 109, in main
    allow_sample_reuse)
  File "create_tf_record_from_joint_dataset.py", line 141, in attempt_to_find_sample_that_is_not_yet_in_dataset
    return random.choice(all_items_in_category)
  File "/opt/python/3.7.15/lib/python3.7/random.py", line 261, in choice
    raise IndexError('Cannot choose from an empty sequence') from None
IndexError: Cannot choose from an empty sequence

When list of documents in given engraving is empty, the random.choice crashes. The code assumes that there will be some documents, even though in documentation it is said, that engraving specifications can be left out.

How to replicate

1. create dataset with no information about engravings
2. run create_tf_record_from_joint_dataset.py with the dataset created above

Possible fix

# create_tf_record_from_joint_dataset.py, line 97
with tqdm(desc="Serializing annotations", total=target_size) as progress_bar:
        while samples_written < target_size:                
            current_engraving, number_of_staves = sampling_categories[index % len(sampling_categories)]
            all_items_in_category = dataset[current_engraving][number_of_staves]

            # fix, check if there are any documents
            if len(all_items_in_category) == 0:
                index += 1
                continue
            # end fix

            encoding_succeeded = False
            tf_example = None

[apacha]

If you've identified the solution, would you mind opening a PR? I'm happy to merge the fix.