nlp.update() 在空白和预训练 SpaCy 模型上是否有区别？答案

【问题标题】：Is there a difference while nlp.update() on blank and pretrained SpaCy models?nlp.update() 在空白和预训练 SpaCy 模型上是否有区别？
【发布时间】：2021-08-30 12:37:20
【问题描述】：

我有一个带注释的数据集 (TRAIN_DATA)，用于构建我自己的 NER 模型：

nlp = spacy.blank("en")

if "ner" not in nlp.pipe_names:
    nlp.add_pipe("ner", last=True)

examples_train = []
for text, annotations in TRAIN_DATA:
    examples_train.append(Example.from_dict(nlp.make_doc(text)

pipe_exceptions = ["ner"]
other_pipes = [pipe for pipe in nlp.pipe_names if pipe not in pipe_exceptions]

with nlp.disable_pipes(*other_pipes):
    if model is None:
        optimizer_default = nlp.initialize()
    else:
        nlp.create_optimizer()

    for itn in range(nIter):
        random.shuffle(examples_train)
        losses_train = {}
        batches = minibatch(examples_train, size=compounding(4.0, 32.0, 1.001))
        for batch in batches:
            try:
                if model is None:
                    nlp.update(
                        batch,
                        drop=dropout,
                        losses=losses_train,
                        sgd=optimizer_default,
                    )
                else:
                    nlp.update(
                        batch,
                        drop=dropout,
                        losses=losses_train
                    )

此代码在创建空白模型时运行良好，但是在尝试更新现有 en_core_web_trf 模型时，我收到了 ValueError，请参阅完整跟踪：

train_model(model, os.path.dirname(os.path.abspath(__file__)) + '/trained_models/' + modelFile, useCuda, spacy_model_type)
File "/Users/miloscuculovic/PycharmProjects/NER_models_reviewer_comments/main.py", line 28, in train_model
nlp, plt = trainSpacyModel(path_train_data, path_valid_data, LABEL, dropout, nIter, spacy_model_type)
File "/Users/miloscuculovic/PycharmProjects/NER_models_reviewer_comments/dospacy.py", line 283, in trainSpacyModel
nlp, plt = trainSpacy(TRAIN_DATA, VALID_DATA, dropout, nIter, spacy_model_type)
File "/Users/miloscuculovic/PycharmProjects/NER_models_reviewer_comments/dospacy.py", line 186, in trainSpacy
nlp.update(
File "/Users/miloscuculovic/PycharmProjects/NER_models_reviewer_comments/venv/lib/python3.8/site-packages/spacy/language.py", line 1123, in update
proc.update(examples, sgd=None, losses=losses, **component_cfg[name])
File "spacy/pipeline/transition_parser.pyx", line 395, in spacy.pipeline.transition_parser.Parser.update
File "/Users/miloscuculovic/PycharmProjects/NER_models_reviewer_comments/venv/lib/python3.8/site-packages/thinc/model.py", line 309, in begin_update
return self._func(self, X, is_train=True)
File "/Users/miloscuculovic/PycharmProjects/NER_models_reviewer_comments/venv/lib/python3.8/site-packages/spacy/ml/tb_framework.py", line 33, in forward
step_model = ParserStepModel(
File "spacy/ml/parser_model.pyx", line 216, in spacy.ml.parser_model.ParserStepModel.__init__
File "/Users/miloscuculovic/PycharmProjects/NER_models_reviewer_comments/venv/lib/python3.8/site-packages/thinc/model.py", line 291, in __call__
return self._func(self, X, is_train=is_train)
File "/Users/miloscuculovic/PycharmProjects/NER_models_reviewer_comments/venv/lib/python3.8/site-packages/thinc/layers/chain.py", line 54, in forward
Y, inc_layer_grad = layer(X, is_train=is_train)
File "/Users/miloscuculovic/PycharmProjects/NER_models_reviewer_comments/venv/lib/python3.8/site-packages/thinc/model.py", line 291, in __call__
return self._func(self, X, is_train=is_train)
File "/Users/miloscuculovic/PycharmProjects/NER_models_reviewer_comments/venv/lib/python3.8/site-packages/thinc/layers/chain.py", line 54, in forward
Y, inc_layer_grad = layer(X, is_train=is_train)
File "/Users/miloscuculovic/PycharmProjects/NER_models_reviewer_comments/venv/lib/python3.8/site-packages/thinc/model.py", line 291, in __call__
return self._func(self, X, is_train=is_train)
File "/Users/miloscuculovic/PycharmProjects/NER_models_reviewer_comments/venv/lib/python3.8/site-packages/spacy_transformers/layers/listener.py", line 58, in forward
model.verify_inputs(docs)
File "/Users/miloscuculovic/PycharmProjects/NER_models_reviewer_comments/venv/lib/python3.8/site-packages/spacy_transformers/layers/listener.py", line 47, in verify_inputs
raise ValueError
ValueError

【问题讨论】：

标签： machine-learning spacy training-data named-entity-recognition spacy-3

【解决方案1】：

看来我的问题与 SpaCy Github 上提出的以下问题有关：https://github.com/explosion/spaCy/issues/6675

通过添加'transformer' 更改pipe_exceptions 解决了这个问题。

所以，我改变了：

pipe_exceptions = ["ner"]

到

pipe_exceptions = ["ner", "transformer"]

【讨论】：