CODEWITHSUNDEEP
transformer-modelcode-documentationspacy-3spacy-transformers
Michael Wade
Michael Wade

Reputation: 232

Spacy 3.1 example code for Transformers in the online documentation seems to be wrong

In the latest documentation for Spacy, the following example is given at the following link:

https://spacy.io/usage/embeddings-transformers

import spacy
from spacy.tokens import Doc
from spacy_transformers import TransformerData
from thinc.api import set_gpu_allocator, require_gpu


def custom_annotation_setter(docs, trf_data):
    doc_data = list(trf_data.doc_data)
    for doc, data in zip(docs, doc_data):
        doc._.custom_attr = data


nlp = spacy.load("en_core_web_trf")
nlp.get_pipe("transformer").set_extra_annotations = custom_annotation_setter
doc = nlp("This is a text")
assert isinstance(doc._.custom_attr, TransformerData)
print(doc._.custom_attr.tensors)

This code throws an exception when it try's to process the test data:

AttributeError: [E047] Can't assign a value to unregistered extension attribute 'custom_attr'. Did you forget to call the set_extension method?

I set the extension using:

Doc.set_extension('custom_attr', default=True)

My question is, should the Transform class handle adding this special extension itself (as is implied in the example code), or is this just a bug in the example?

Upvotes: 1

Views: 2205

Answers (1)

polm23
polm23

Reputation: 15623

Your code runs without error for me if I set the extension before your function definition.

import spacy
from spacy.tokens import Doc
from spacy_transformers import TransformerData
from thinc.api import set_gpu_allocator, require_gpu

Doc.set_extension('custom_attr', default=True)

def custom_annotation_setter(docs, trf_data):
    doc_data = list(trf_data.doc_data)
    for doc, data in zip(docs, doc_data):
        doc._.custom_attr = data


nlp = spacy.load("en_core_web_trf")
nlp.get_pipe("transformer").set_extra_annotations = custom_annotation_setter
doc = nlp("This is a text")
assert isinstance(doc._.custom_attr, TransformerData)
print(doc._.custom_attr.tensors)

Maybe you called the set_extension in the wrong place?

Upvotes: 1

Related Questions