Datasets:

pie

/

aae2

like 0

Follow

pie 3

Hi @taghizadeh ,

I implemented this fix: https://huggingface.co/datasets/pie/aae2/discussions/2

Can you try if this works for you, i.e. try passing revision="pr/2" to load_dataset?

Yes thank you. The load_dataset works fine right now, but when trying to access the samples in train or test section I get the following error:

sample = dataset['train'][0]

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
Cell In[8], line 1
----> 1 dataset['test'][0]

File ~/Library/Caches/pypoetry/virtualenvs/minig-Z580gx8g-py3.11/lib/python3.11/site-packages/datasets/arrow_dataset.py:2795, in Dataset.__getitem__(self, key)
   2793 def __getitem__(self, key):  # noqa: F811
   2794     """Can be used to index columns (by string names) or rows (by integer index or iterable of indices or bools)."""
-> 2795     return self._getitem(key)

File ~/Library/Caches/pypoetry/virtualenvs/minig-Z580gx8g-py3.11/lib/python3.11/site-packages/datasets/arrow_dataset.py:2780, in Dataset._getitem(self, key, **kwargs)
   2778 formatter = get_formatter(format_type, features=self._info.features, **format_kwargs)
   2779 pa_subtable = query_table(self._data, key, indices=self._indices if self._indices is not None else None)
-> 2780 formatted_output = format_table(
   2781     pa_subtable, key, formatter=formatter, format_columns=format_columns, output_all_columns=output_all_columns
   2782 )
   2783 return formatted_output

File ~/Library/Caches/pypoetry/virtualenvs/minig-Z580gx8g-py3.11/lib/python3.11/site-packages/datasets/formatting/formatting.py:629, in format_table(table, key, formatter, format_columns, output_all_columns)
    627 python_formatter = PythonFormatter(features=formatter.features)
    628 if format_columns is None:
--> 629     return formatter(pa_table, query_type=query_type)
    630 elif query_type == "column":
    631     if key in format_columns:

File ~/Library/Caches/pypoetry/virtualenvs/minig-Z580gx8g-py3.11/lib/python3.11/site-packages/datasets/formatting/formatting.py:396, in Formatter.__call__(self, pa_table, query_type)
    394 def __call__(self, pa_table: pa.Table, query_type: str) -> Union[RowFormat, ColumnFormat, BatchFormat]:
    395     if query_type == "row":
--> 396         return self.format_row(pa_table)
    397     elif query_type == "column":
    398         return self.format_column(pa_table)

File ~/Library/Caches/pypoetry/virtualenvs/minig-Z580gx8g-py3.11/lib/python3.11/site-packages/pie_datasets/core/document_formatter.py:15, in DocumentFormatter.format_row(self, pa_table)
     13 def format_row(self, pa_table: pa.Table) -> Document:
     14     row = self.python_arrow_extractor().extract_row(pa_table)
---> 15     return self.document_type.fromdict(row)

File ~/Library/Caches/pypoetry/virtualenvs/minig-Z580gx8g-py3.11/lib/python3.11/site-packages/pytorch_ie/core/document.py:719, in Document.fromdict(cls, dct)
    717 annotation_id = annotation_dict.pop("_id")
    718 # annotations can only reference annotations
--> 719 annotation = annotation_class.fromdict(annotation_dict, annotations)
    720 annotations[annotation_id] = annotation
    721 annotations_per_field[field.name].append(annotation)

File ~/Library/Caches/pypoetry/virtualenvs/minig-Z580gx8g-py3.11/lib/python3.11/site-packages/pytorch_ie/core/document.py:306, in Annotation.fromdict(cls, dct, annotation_store)
    303         raise Exception(f"unknown annotation container type: {container_type}")
    305 tmp_dct.pop("_id", None)
--> 306 return cls(**tmp_dct)

TypeError: LabeledSpan.__init__() got an unexpected keyword argument 'slices'

ok, I added another PR with a fix: https://huggingface.co/datasets/pie/aae2/discussions/3.
can you try with revision="pr/3"?

Sorry for delay, Yep it works now.

Another issue. When I use connect_first or connect_all, the relations of claims are not appended to premises relations as I see in the code.

What exactly do you mean? Can you give me a pointer to the respective lines of code and explain what you expect instead?

Well, we expect if we use use connect_first argument, Claims be connected to first MajorClaim, but when getting the relations from a document I only get the relations between premises and claims. This is the same with semantically same relations.

The line of related code:
https://huggingface.co/datasets/pie/aae2/blob/1015ee38bd8a36549b344008f7a49af72956a7fe/aae2.py#L79

for example I loaded the first document it has 6 relations with 11 spans. For 11 spans there must be at least 10 relations if consider all of them connected.

I assume you did not trigger the document conversion. The idea is that when using load_dataset(), the data is loaded as it is without any semantic changes. Just when converting it, either by calling dataset.to_document_type(ONE_OF_THE_KEYS_IN_THE_DOCUMENT_CONVERTERS)(see in the documentation or the source code for the predefined converters) or dataset.map(function), the semantics should change (e.g. new relations are added).

from pie_datasets import load_dataset, DatasetDict
from pytorch_ie.documents import TextDocumentWithLabeledSpansAndBinaryRelations

dataset: DatasetDict = load_dataset("pie/aae2", conversion_method="connect_all", revision="pr/3")
# convert the dataset. this will use either connect_all or connect_first method variant to connect all claims
dataset_converted = dataset.to_document_type(TextDocumentWithLabeledSpansAndBinaryRelations)
doc: TextDocumentWithLabeledSpansAndBinaryRelations = dataset_converted["train"][0]
print(len(doc.binary_relations))  # 12
print(len(doc.labeled_spans))  # 11

# get resolved relations for better debugging (requires pytorch-ie>=0.30.2)
resolved_relations = doc.binary_relations.resolve()
print(resolved_relations)

Does this help?

You are right. My apologies, for not reading the documentation well.

cool :) I merged https://huggingface.co/datasets/pie/aae2/discussions/3, do you think we can close this?

Yes thank you.