BERT-BiLSMT-CRF-NER

Tensorflow solution of NER task Using BiLSTM-CRF model with Google BERT Fine-tuning

使用谷歌的BERT模型在BLSTM-CRF模型上进行预训练用于中文命名实体识别的Tensorflow代码'

中文文档请查看https://blog.csdn.net/macanv/article/details/85684284 如果对您有帮助，麻烦点个star,谢谢~~

Welcome to star this repository!

The Chinese training data($PATH/NERdata/) come from:https://github.com/zjy-ucas/ChineseNER

The CoNLL-2003 data($PATH/NERdata/ori/) come from:https://github.com/kyzhouhzau/BERT-NER

The evaluation codes come from:https://github.com/guillaumegenthial/tf_metrics/blob/master/tf_metrics/__init__.py

Try to implement NER work based on google's BERT code and BiLSTM-CRF network! This project may be more close to process Chinese data. but other language only need Modify a small amount of code.

THIS PROJECT ONLY SUPPORT Python3.
###################################################################

Download project and install

You can install this project by:

pip install bert-base==0.0.5 -i https://pypi.python.org/simple

OR

git clone https://github.com/macanv/BERT-BiLSTM-CRF-NER
cd BERT-BiLSTM-CRF-NER/
python3 setup.py install

UPDATE:

1. fix Missing loss error
2. add label_list params in train process, so you can using -label_list xxx to special labels in training process.

Train model:

You can use -help to view the relevant parameters of the training named entity recognition model, where data_dir, bert_config_file, output_dir, init_checkpoint, vocab_file must be specified.

bert-base-ner-train -help

train/dev/test dataset is like this:

海 O
钓 O
比 O
赛 O
地 O
点 O
在 O
厦 B-LOC
门 I-LOC
与 O
金 B-LOC
门 I-LOC
之 O
间 O
的 O
海 O
域 O
。 O

The first one of each line is a token, the second is token's label, and the line is divided by a blank line. The maximum length of each sentence is [max_seq_length] params.
You can get training data from above two git repos
You can training ner model by running below command:

bert-base-ner-train \
    -data_dir {your dataset dir}\
    -output_dir {training output dir}\
    -init_checkpoint {Google BERT model dir}\
    -bert_config_file {bert_config.json under the Google BERT model dir} \
    -vocab_file {vocab.txt under the Google BERT model dir}

you can special labels using -label_list params, the project get labels from training data.

# using , split
-labels 'B-LOC, I-LOC ...'
OR save label in a file like labels.txt, one line one label
-labels labels.txt

After training model, the NER model will be saved in {output_dir} which you special above cmd line.

As Service

Many server and client code comes from excellent open source projects: bert as service of hanxiao If my code violates any license agreement, please let me know and I will correct it the first time. and NER server/client service code can be applied to other tasks with simple modifications, such as text categorization, which I will provide later. Welcome to submit your request, if you want to share it on Github or my work.

You can use -help to view the relevant parameters of the NER as Service: which ner_model_dir, bert_model_dir is need

bert-base-serving-start -help

and than you can using below cmd start ner service:

bert-base-serving-start \
    -ner_model_dir C:\workspace\python\BERT_Base\output\ner2 \
    -bert_model_dir F:\chinese_L-12_H-768_A-12
    -model_pb_dir C:\workspace\python\BERT_Base\model_pb_dir

as you see:
mode: If mode is NER, then the service identified by the named entity will be started. If it is BERT, it will be the same as the [bert as service] project.
bert_model_dir: bert_model_dir is a BERT model, you can download from https://github.com/google-research/bert ner_model_dir: your ner model checkpoint dir model_pb_dir: model freeze save dir, after run optimize func, there will contains like ner_model.pb binary file

You can download my ner model from：https://pan.baidu.com/s/1m9VcueQ5gF-TJc00sFD88w, ex_code: guqq
Set ner_mode.pb to model_pb_dir, and set other file to ner_model_dir

You can see below service starting info:

you can using below code test client:

import time
from bert_base.client import BertClient

with BertClient(show_server_config=False, check_version=False, check_length=False, mode='NER') as bc:
    start_t = time.perf_counter()
    str = '1月24日，新华社对外发布了中央对雄安新区的指导意见，洋洋洒洒1.2万多字，17次提到北京，4次提到天津，信息量很大，其实也回答了人们关心的很多问题。'
    rst = bc.encode([str, str])
    print('rst:', rst)
    print(time.perf_counter() - start_t)

you can see this after run the above code:

The following tutorial is an old version and will be removed in the future.

How to train

1. Download BERT chinese model :

wget https://storage.googleapis.com/bert_models/2018_11_03/chinese_L-12_H-768_A-12.zip  

2. create output dir

create output path in project path:

mkdir output

3. Train model

first method

  python3 bert_lstm_ner.py   \
                  --task_name="NER"  \ 
                  --do_train=True   \
                  --do_eval=True   \
                  --do_predict=True
                  --data_dir=NERdata   \
                  --vocab_file=checkpoint/vocab.txt  \ 
                  --bert_config_file=checkpoint/bert_config.json \  
                  --init_checkpoint=checkpoint/bert_model.ckpt   \
                  --max_seq_length=128   \
                  --train_batch_size=32   \
                  --learning_rate=2e-5   \
                  --num_train_epochs=3.0   \
                  --output_dir=./output/result_dir/ 

OR replace the BERT path and project path in bert_lstm_ner.py

if os.name == 'nt': #windows path config
   bert_path = '{your BERT model path}'
   root_path = '{project path}'
else: # linux path config
   bert_path = '{your BERT model path}'
   root_path = '{project path}'

Than Run:

python3 bert_lstm_ner.py

USING BLSTM-CRF OR ONLY CRF FOR DECODE!

Just alter bert_lstm_ner.py line of 450, the params of the function of add_blstm_crf_layer: crf_only=True or False

ONLY CRF output layer:

    blstm_crf = BLSTM_CRF(embedded_chars=embedding, hidden_unit=FLAGS.lstm_size, cell_type=FLAGS.cell, num_layers=FLAGS.num_layers,
                          dropout_rate=FLAGS.droupout_rate, initializers=initializers, num_labels=num_labels,
                          seq_length=max_seq_length, labels=labels, lengths=lengths, is_training=is_training)
    rst = blstm_crf.add_blstm_crf_layer(crf_only=True)

BiLSTM with CRF output layer

    blstm_crf = BLSTM_CRF(embedded_chars=embedding, hidden_unit=FLAGS.lstm_size, cell_type=FLAGS.cell, num_layers=FLAGS.num_layers,
                          dropout_rate=FLAGS.droupout_rate, initializers=initializers, num_labels=num_labels,
                          seq_length=max_seq_length, labels=labels, lengths=lengths, is_training=is_training)
    rst = blstm_crf.add_blstm_crf_layer(crf_only=False)

Result:

all params using default

In dev data set:

In test data set

entity leval result:

last two result are label level result, the entitly level result in code of line 796-798,this result will be output in predict process. show my entity level result :

my model can download from baidu cloud:
链接：https://pan.baidu.com/s/1GfDFleCcTv5393ufBYdgqQ 提取码：4cus
NOTE: My model is trained by crf_only params

ONLINE PREDICT

If model is train finished, just run

python3 terminal_predict.py

Using NER as Service

Service

Using NER as Service is simple, you just need to run the python script below in the project root path:

python3 runs.py \ 
    -mode NER
    -bert_model_dir /home/macan/ml/data/chinese_L-12_H-768_A-12 \
    -ner_model_dir /home/macan/ml/data/bert_ner \
    -model_pd_dir /home/macan/ml/workspace/BERT_Base/output/predict_optimizer \
    -num_worker 8

You can download my ner model from：https://pan.baidu.com/s/1m9VcueQ5gF-TJc00sFD88w, ex_code: guqq
Set ner_mode.pb to model_pd_dir, and set other file to ner_model_dir and than run last cmd

Client

The client using methods can reference client_test.py script

import time
from client.client import BertClient

ner_model_dir = 'C:\workspace\python\BERT_Base\output\predict_ner'
with BertClient( ner_model_dir=ner_model_dir, show_server_config=False, check_version=False, check_length=False, mode='NER') as bc:
    start_t = time.perf_counter()
    str = '1月24日，新华社对外发布了中央对雄安新区的指导意见，洋洋洒洒1.2万多字，17次提到北京，4次提到天津，信息量很大，其实也回答了人们关心的很多问题。'
    rst = bc.encode([str])
    print('rst:', rst)
    print(time.perf_counter() - start_t)

NOTE: input format you can sometime reference bert as service project.
Welcome to provide more client language code like java or others.

Using yourself data to train

if you want to use yourself data to train ner model,you just modify the get_labes func.

def get_labels(self):
       return ["O", "B-PER", "I-PER", "B-ORG", "I-ORG", "B-LOC", "I-LOC", "X", "[CLS]", "[SEP]"]

NOTE: "X", “[CLS]”, “[SEP]” These three are necessary, you just replace your data label to this return list.
Or you can use last code lets the program automatically get the label from training data

def get_labels(self):
        # 通过读取train文件获取标签的方法会出现一定的风险。
        if os.path.exists(os.path.join(FLAGS.output_dir, 'label_list.pkl')):
            with codecs.open(os.path.join(FLAGS.output_dir, 'label_list.pkl'), 'rb') as rf:
                self.labels = pickle.load(rf)
        else:
            if len(self.labels) > 0:
                self.labels = self.labels.union(set(["X", "[CLS]", "[SEP]"]))
                with codecs.open(os.path.join(FLAGS.output_dir, 'label_list.pkl'), 'wb') as rf:
                    pickle.dump(self.labels, rf)
            else:
                self.labels = ["O", 'B-TIM', 'I-TIM', "B-PER", "I-PER", "B-ORG", "I-ORG", "B-LOC", "I-LOC", "X", "[CLS]", "[SEP]"]
        return self.labels

NEW UPDATE

2019.1.30 Support pip install and command line control

2019.1.30 Add Service/Client for NER process

2019.1.9: Add code to remove the adam related parameters in the model, and reduce the size of the model file from 1.3GB to 400MB.

2019.1.3: Add online predict code

reference:

• The evaluation codes come from:https://github.com/guillaumegenthial/tf_metrics/blob/master/tf_metrics/__init__.py

• https://github.com/google-research/bert

• https://github.com/kyzhouhzau/BERT-NER

• https://github.com/zjy-ucas/ChineseNER

• https://github.com/hanxiao/bert-as-service

Any problem please open issue OR email me(ma_cancan@163.com)