BILSTM-CRF命名实体识别—keras版代码详解

本文代码源自:https://www.cnblogs.com/vipyoumay/p/ner-chinese-keras.html

github:https://github.com/stephen-v/zh-NER-keras

运行步骤:

1.首先要安装GIT

2.安装 keras-contrib:pip install git+https://www.github.com/keras-team/keras-contrib.git

3.运行,报错行:import bilsm_crf_model 错误类型:ImportError: cannot import name normalize_data_format

方法:https://github.com/ekholabs/keras-contrib/commit/0dac2da8a19f34946448121c6b9c8535bfb22ce2 

4.运行又报错:错误类型:AttributeError: ‘CRF’ object has no attribute ‘_inbound_nodes’

方法:将keras版本改至2.1.4

5.继续报错:'Tensor' object has no attribute 'assign'

方法:把process_data.py文件中的

    if platform.system() == 'Windows':
        split_text = '\r\n'
    else:
        split_text = '\n'
改为
    if platform.system() == 'Windows':
        split_text = '\n'
    else:
        split_text = '\r\n'

成功运行!

代码详解:

代码错误:(配置问题,配置好了不会错)

word_counts = Counter(row[0].lower() for sample in train for row in sample)
改为
train = train[0][0]
word_counts = Counter(train)

 

你可能感兴趣的:(keras)