Deep Bayesian Semi-Supervised Active Learning for Sequence Labelling

This repository contains experiments for article: Deep Bayesian Semi-Supervised Active Learning for Sequence Labelling.

Getting started

1. Install environment and activate it

   conda env update & source activate deep-bayes-active-semisup

2. Download Glove embeddings and store it in:

   datasets/glove/glove.840B.300d.txt

3. Run experiment

• BI-LSTM-FCN - experiment with unlimited number of tokens

    python -u experiments/bilstm_fcn_unlimited  NER    
    python -u experiments/bilstm_fcn_unlimited  POS
    python -u experiments/bilstm_fcn_unlimited  CHUNK

• BI-LSTM-FCN - experiment with limited number of tokens

    python -u experiments/bilstm_fcn_limited  NER    
    python -u experiments/bilstm_fcn_limited  POS
    python -u experiments/bilstm_fcn_limited  CHUNK

• BI-LSTM-CRF - experiment with unlimited number of tokens

    python -u experiments/bilstm_crf_unlimited  NER    
    python -u experiments/bilstm_crf_unlimited  POS
    python -u experiments/bilstm_crf_unlimited  CHUNK

• BI-LSTM-CRF - experiment with limited number of tokens

    python -u experiments/bilstm_crf_limited  NER    
    python -u experiments/bilstm_crf_limited  POS
    python -u experiments/bilstm_crf_limited  CHUNK