Location>code7788 >text

Training Named Entity Recognition Models with DeepKE DEMO (Official DEMO)

Popularity：298 ℃/2024-10-10 17:08:18

Training Named Entity Recognition Models with DeepKE DEMO (Official DEMO)

Description:

  • First published: 2024-10-10
  • DeepKE Resources:
    • Documentation:/DeepKE/
    • Website:/
    • cnschema： /

Set up Github mirrors if needed

git config --system url."/".insteadOf /

To unset, enter:
git config --system --unset :///.insteadof

Creating a conda environment

conda create -n deepke python=3.8
conda activate deepke

# mountingtorch
pip install torch==1.11.0+cu113 torchvision==0.12.0+cu113 torchaudio==0.11.0 --extra-index-url /whl/cu113

# 使用阿里云镜像mountingtorch 1.11.0
# pip install /pytorch-wheels/cu113/torch-1.11.0+cu113-cp38-cp38-linux_x86_64.whl /pytorch-wheels/cu113/torchvision-0.12.0+cu113-cp38-cp38-linux_x86_64.whl /pytorch-wheels/cu113/torchaudio-0.11.0+cu113-cp38-cp38-linux_x86_64.whl -i /pypi/simple/

Install DeepKE:

git clone /zjunlp/
cd DeepKE

pip install pip==24.0

pip install -r  -i /pypi/simple/
python  install
python  develop

pip install prettytable==2.4.0
pip install ipython==8.12.0

Download Dataset

# apt-get install wget
cd example/ner/standard
wget 120.27.214.45/Data/ner/standard/
tar -xzvf

can be seendataIt's in the folder:

  • : Training set
  • : Validation set
  • : Test set

Configuring wandb

exist/ Register an account on and create a new project with a name like:deepke-ner-official-demo

show (a ticket)/authorize Get API key

(of a computer) runwandb initInput the API key you just obtained and the project you created.

Run training and prediction

Delete the checkpoints and logs folders (if any) saved from previous training:

rm -r checkpoints/
rm -r logs/

lstmcrf

show (a ticket)example/ner/standard/run_lstmcrf.py, make sure the wandb and yaml libraries are imported properly:

import wandb
import yaml

Modify the project name of wandb:

if config['use_wandb']:
    (project="deepke-ner-official-demo")

modificationsexample/ner/standard/conf/ hit the nail on the headuse_wandb because ofTrue。

If you need to train with multiple GPUs, modify theexample/ner/standard/conf/ hit the nail on the headuse_multi_gpu because ofTrue

Start training:

python run_lstmcrf.py

>> total: 109870 loss: 27.181508426008552
              precision    recall  f1-score   support

       B-LOC     0.8920    0.8426    0.8666      1951
       B-ORG     0.8170    0.7439    0.7787       984
       B-PER     0.8783    0.8167    0.8464       884
       I-LOC     0.8650    0.8264    0.8453      2581
       I-ORG     0.8483    0.8365    0.8424      3945
       I-PER     0.8860    0.8436    0.8643      1714
           O     0.9861    0.9912    0.9886     97811

    accuracy                         0.9732    109870
   macro avg     0.8818    0.8430    0.8618    109870
weighted avg     0.9727    0.9732    0.9729    109870

The prediction text used is saved in theexample/ner/standard/conf/in the program, change it to the following:

text: ""Water heaters and other trade-ins saved more than 2,000 yuan."" On Oct. 3, Jin Yu, a citizen, touches his cell phone to place an order, make a payment and register at a shopping plaza in Xiangyang, Hubei province. Hubei is focusing on promoting large-scale equipment renewal and consumer goods trade-in. "Strive to the end of this year, the province's automobile scrapping and renewal, replacement and renewal of 45,000, 125,000, respectively, home appliances to 1.7 million sets of old for new." Long Xiaohong, director of the Hubei Provincial Department of Commerce, introduced."

Operational Forecasting:

python

NER Results.

[('Lake', 'B-LOC'), ('North', 'I-LOC'), ('Province', 'I-LOC'), ('Xiang', 'B-LOC'), ('Yang', 'I-LOC'), ('Municipality', 'I-LOC'), ('Market', 'I-LOC'), ('Yuh', 'I-PER'), ('Lake', 'B-ORG'), ('North ', 'I-ORG'), ('Province', 'I-ORG'), ('Business', 'I-ORG'), ('Service', 'I-ORG'), ('Office', 'I-ORG'), ('Hall', 'I-ORG'), ('Dragon', 'B-PER'), ('Small', 'I-PER'), ('Red', 'I-PER')]

bert

modificationsexample/ner/standard/conf/hit the nail on the headhydra/modelbecause ofbert。

The hyperparameters of bert are set in theexample/ner/standard/conf/hydra/model/, which can be modified if necessary.

modificationsexample/ner/standard/conf/ hit the nail on the headuse_wandb because ofTrue。

modificationsexample/ner/standard/run_bert.py The name of the project in the wandb:

    if cfg.use_wandb:
        (project="deepke-ner-official-demo")

Modify as necessaryexample/ner/standard/conf/hit the nail on the headtrain_batch_sizeFor bert it is recommended to be no smaller than 64

Start training:

export HF_ENDPOINT=
python run_bert.py

w2ner

w2ner is a new SOTA model.

on the basis ofW2NER (AAAI'22) for entity recognition for multiple scenarios (for details, please refer to paperUnified Named Entity Recognition as Word-Word Relation Classification).

Named Entity Recognition (NER) involves three main types of NER, including planar, overlapping (aka nested), and discontinuous NER, which have mostly been studied separately. Recently, there has been an increasing interest in unifying NERs, theW2NERUse one model to handle all three of these tasks simultaneously.

Due to the use of a single card GPU, modifying theexample/ner/standard/w2ner/conf/hit the nail on the headdevice because of0。

modificationsexample/ner/standard/w2ner/conf/hit the nail on the headdata_dircap (a poem)do_train：

data_dir: "../data"
do_train: True

in order to use the previously downloaded dataset and start training.

Run training:

python
<Previous Next>