NER TF 2.0 inference(single sentence)for CoNLL-2003 NER dataset.

A detailed blog for training the model in TF 2.0 is available at

https://medium.com/analytics-vidhya/ner-tensorflow-2-2-0-9f10dcf5a0a.

Folder structure

• Download and store glove.6B.100d.txt from here
• data folder has the dataset.
• model output files after training will be written to the model_output directory.

├── data
│   ├── train.txt
│   ├── valid.txt
│   ├── test.txt
│
├── embeddings
│   ├── glove.6B.100d.txt
│   
├── model_output
│   ├── checkpoint
│   ├── embedding.pkl
│   ├── idx2Label.pkl
│   ├── model_weights.data-00000-of-00001
│   ├── model_weights.index
│   ├── word2Idx.pkl
│   

Setup

python3 -m venv venv1
source venv1/bin/activate
pip install -r requirements.txt

Model training

python3 train.py --data data --output  model_output --overwrite True 

API

To use api.py uncomment the

python3 api.py

WSGI server (gunicorn with nginx)

gunicorn --bind :8088 --workers 1 --threads 8 api:app

cURL

• test

curl -i -H "Content-Type: application/json" \
-X POST http://0.0.0.0:8088/test

• single sentence

curl -i -H "Content-Type: application/json" \
-X POST http://0.0.0.0:8088/ner -d '{"sent":"Steve went to Paris"}'

Visualisations

tensorboard --logdir=model_output/logs/train --port=6006 --bind_all
tensorboard --logdir=model_output/logs/valid --port=6006 --bind_all