Spacy 3中命名实体识别的置信度分数

Question

我需要获取由NER 'de_core_news_lg'模型预测的标签的置信度分数。在Spacy 2中，已经有一个众所周知的解决方案：

nlp = spacy.load('de_core_news_lg')
doc = nlp('ich möchte mit frau Mustermann in der Musterbank sprechen')
text = content
doc = nlp.make_doc(text)
beams = nlp.entity.beam_parse([doc], beam_width=16, beam_density=0.0001)
for score, ents in nlp.entity.moves.get_beam_parses(beams[0]):
    print (score, ents)
    entity_scores = defaultdict(float)
    for start, end, label in ents:
        # print ("here")
        entity_scores[(start, end, label)] += score
        print ('entity_scores', entity_scores)

然而，在Spacy 3中我遇到了以下错误:

AttributeError: 'German' object has no attribute 'entity'

显然，language对象不再具有entity属性。有人知道如何在Spacy 3中获取置信度分数吗？

- Keyvan Sadri

---

这是一个重复的问题，链接为 https://stackoverflow.com/questions/67421308/spacy-3-beam-parse-for-ner-probability - polm23

Answer 1

答案的核心是“使用管道组件“beam_ner”，并查看EntityRecognizer.pyx代码。然后有单元测试test_ner.py test_beam_ner_scores()，几乎展示了如何做到这一点。如果您想查看如何修改配置文件cfg，请保存模型（如下所示make_nlp()），并查看保存的模型config.cfg。

问题在于它仅适用于单元测试生成的“model”。对于我的真实模型（5000个文档，每个文本约4k，训练NER F得分约为75％），它表现不佳。当我说“表现不佳”时，我是指“贪婪搜索”可以找到我的实体，但“波束搜索”会报告数百个标记（甚至包括标点符号），其“分数”通常为0.013。根据偏移量，这些标记通常来自文档的一个小部分。

这很令人沮丧，因为我相信spacy train（针对'beam_ner'）使用相同的代码来“验证”训练迭代，并且训练报告的得分几乎还不错（好吧，比Spacy 2低10％，但这也发生在使用'ner'和'beam_ner'进行训练时）。

因此，我发布这篇文章，希望有人有更好的运气或指出我做错了什么。

到目前为止，Spacy3 对我来说是一个重大灾难：无法获得置信度，也不能再使用 GPU（我只有 6GB），基于 Ray 的并行化不起作用（在 Windows 上 = 实验性质），而使用“transformer”模型进行实体命名识别的训练分数比 Spacy 2 差了 10%。

代码

import spacy
from spacy.lang.en import English
from spacy.language import Language
from spacy.tokens import Doc
from spacy.training import Example

# Based upon test_ner.py test_beam_ner_scores()

TRAIN_DATA = [
    ("Who is Shaka Khan?", {"entities": [(7, 17, "PERSON")]}),
    ("I like London and Berlin.",  {"entities": [(7, 13, "LOC"), (18, 24, "LOC")]}),
    ("You like Paris and Prague.", {"entities": [(9, 14, "LOC"), (19, 25, "LOC")]}),
]

def make_nlp(model_dir):
    # ORIGINALLY: Test that we can get confidence values out of the beam_ner pipe
    nlp = English()
    config = { "beam_width": 32, "beam_density": 0.001 }
    ner = nlp.add_pipe("beam_ner", config=config)
    train_examples = []
    for text, annotations in TRAIN_DATA:
        train_examples.append(Example.from_dict(nlp.make_doc(text), annotations))
        for ent in annotations.get("entities"):
            ner.add_label(ent[2])
    optimizer = nlp.initialize()
    # update once
    losses = {}
    nlp.update(train_examples, sgd=optimizer, losses=losses)
    # save
    #if not model_dir.exists():
    #model_dir.mkdir()
    nlp.to_disk(model_dir)
    print("Saved model to", model_dir)
    return nlp


def test_greedy(nlp, text):
    # Report predicted entities using the default 'greedy' search (no confidences)
    doc = nlp(text)    
    print("GREEDY search");
    for ent in doc.ents:
        print("Greedy offset=", ent.start_char, "-", ent.end_char, ent.label_, "text=",  ent.text)
 
def test_beam(nlp, text):
    # Report predicted entities using the beam search (beam_width 16 or higher)
    ner = nlp.get_pipe("beam_ner")

    # Get the prediction scores from the beam search
    doc = nlp.make_doc(text)
    docs = [doc]
    # beams = StateClass returned from ner.predict(docs)
    beams = ner.predict(docs)
    print("BEAM search, labels", ner.labels);

    # Show individual entities and their scores as reported
    scores = ner.scored_ents(beams)[0]
    for ent, sco in scores.items():
        tok = doc[ent[0]]
        lbl = ent[2]
        spn = doc[ent[0]: ent[1]]           
        print('Beam-search', ent[0], ent[1], 'offset=', tok.idx, lbl, 'score=', sco,
              'text=', spn.text.replace('\n', '  '))

MODEL_DIR = "./test_model"
TEST_TEXT = "I like London and Paris."
  
if __name__ == "__main__":
    # You may have to repeat make_nlp() several times to produce a semi-decent 'model'
    # nlp = make_nlp(MODEL_DIR)
    nlp = spacy.load(MODEL_DIR)
    test_greedy(nlp, TEST_TEXT)
    test_beam  (nlp, TEST_TEXT)

结果应该看起来像这样（在重复使用make_nlp生成可用的“模型”之后）：

GREEDY search
Greedy offset= 7 - 13 LOC text= London
Greedy offset= 18 - 23 LOC text= Paris
BEAM search, labels ('LOC', 'PERSON')
Beam-search 2 3 offset= 7 LOC score= 0.5315668466265199 text= London
Beam-search 4 5 offset= 18 LOC score= 0.7206478212662492 text= Paris
Beam-search 0 1 offset= 0 LOC score= 0.4679245513356703 text= I
Beam-search 3 4 offset= 14 LOC score= 0.4670399792743775 text= and
Beam-search 5 6 offset= 23 LOC score= 0.2799470367073933 text= .
Beam-search 1 2 offset= 2 LOC score= 0.21658368070744227 text= like

- mbrunecky

Answer 2

目前在spaCy v3中没有一个好的方法来获取NER分数的置信度。然而，正在开发一个SpanCategorizer组件，将使这个过程变得容易。虽然不确定，但我们希望在下一个小版本中发布它。您可以在功能的PR中跟踪开发，或者在这里阅读更多信息。

- polm23