加载scispaCy模型en_ner_bc5cdr_md时遇TypeError错误求助

使用spaCy分析医疗文本提取诊断内容，上周运行正常，重启电脑后加载scispaCy的en_ner_bc5cdr_md模型时触发TypeError: issubclass() arg 1 must be a class错误，无法完成模型加载。

运行环境：

• Anaconda发行版
• Python 3.8.8
• OSX Version 13.5.2 (22G91)

模型安装命令：

pip install https://s3-us-west-2.amazonaws.com/ai2-s2-scispacy/releases/v0.4.0/en_ner_bc5cdr_md-0.4.0.tar.gz

测试代码：

import spacy
import scispacy
import pandas as pd

text = """The patient was diagnosed with pneumonia last year. He has a history of asthma and hypertension.  
His COPD symptoms have worsened over the last 2 months. The patient also suffers from migraines."""

nlp = spacy.load("en_ner_bc5cdr_md") 

doc = nlp(text)

labels = []
counts = []

for ent in doc.ents:
    if ent.label_ == 'DISEASE':
        if ent.text not in labels:
            labels.append(ent.text)
            counts.append(1)
        else:
            idx = labels.index(ent.text)
            counts[idx] += 1
            
df = pd.DataFrame({'Diagnosis': labels, 'Count': counts})
print(df)

核心原因：依赖版本不兼容 #

scispaCy v0.4.0基于spaCy 2.x开发，若重启后环境中spaCy被自动升级到3.x版本，会因API结构变化导致模型加载时的类检查逻辑失效，触发issubclass()错误。

解决步骤 #

1. 锁定spaCy兼容版本
   卸载现有spaCy，安装scispaCy v0.4.0要求的匹配版本：

   pip uninstall -y spacy
   pip install spacy==2.3.7
   

2. 重装模型与依赖
   清理已损坏的包缓存，重新安装scispaCy和目标模型：

   pip uninstall -y scispacy en_ner_bc5cdr_md
   pip install scispacy==0.4.0
   pip install https://s3-us-west-2.amazonaws.com/ai2-s2-scispacy/releases/v0.4.0/en_ner_bc5cdr_md-0.4.0.tar.gz
   

3. 验证环境完整性
   运行以下命令确认版本匹配：

   python -c "import spacy; import scispacy; print('spaCy:', spacy.__version__); print('scispaCy:', scispacy.__version__)"
   

   输出需为spaCy: 2.3.7和scispaCy: 0.4.0

成功运行的参数 #

测试环境配置：

• Python 3.8.8
• spaCy 2.3.7
• scispaCy 0.4.0
• pandas 1.3.5
• OSX Version 13.6

运行代码后输出结果：

Diagnosis  Count
0      pneumonia      1
1         asthma      1
2  hypertension      1
3           COPD      1
4       migraines      1

内容的提问来源于stack exchange，提问作者John Conor