bi-lstm-crf 0.1.0

A PyTorch implementation of the BI-LSTM-CRF model

A PyTorch implementation of the BI-LSTM-CRF model.

Features:

• General implementation of CRF module
  • Full vectorized implementation: 1) no "for loop" in score sentence algorithm; 2) single level "for loop" in viterbi and forward algorithm
  • START/STOP tags are automatically added in CRF
  • A inner Linear Layer is included which transform from features space to tag space
• General implementation of BI-LSTM-CRF model
• CUDA supported
• Full support for batch computation: training, predicting
• Specialized for NLP sequence tagging tasks
• Easy to train your own sequence tagging models
• MIT License

INSTALL

• dependencies
  • Python 3
  • PyTorch
• install

  $ git clone https://github.com/jidasheng/bi-lstm-crf.git
  $ pip install bi-lstm-crf/
  
  $ pip install bi-lstm-crf   # sometimes later
  

TRAIN

corpus

• prepare your corpus in the specified structure and format.
• there is also a sample corpus in bi_lstm_crf/app/sample_corpus.

train

• training

  $ python -m bi_lstm_crf corpus_dir --model_dir "model_xxx"
  

  • more options
• training results
  • the training results are saved in the model_dir, you can load and draw loss curve

    import pandas as pd
    import matplotlib.pyplot as plt
    
    df = pd.read_csv(".../model_dir/loss.csv")
    df[["train_loss", "val_loss"]].ffill().plot(grid=True)
    plt.show()
    

PREDICT

from bi_lstm_crf.app import WordsTagger

model = WordsTagger(model_dir)
print(model(["市领导到成都..."]))  # CHAR-based model
print(model([["市", "领导", "到", "成都", ...]]))  # WORD-based model

MODULES

There modules are generally built and can be used in other projects

• CRF
• BiRnnCrf

References

• Zhiheng Huang, Wei Xu, and Kai Yu. 2015. Bidirectional LSTM-CRF Models for Sequence Tagging. arXiv:1508.01991.
• Pytorch tutorial ADVANCED: MAKING DYNAMIC DECISIONS AND THE BI-LSTM CRF