Skip to content

Latest commit

 

History

History
32 lines (32 loc) · 1.1 KB

README.md

File metadata and controls

32 lines (32 loc) · 1.1 KB

spoken_summ_div

Requirements (tested)

  • python 3.7
  • torch 1.2
  • CUDA 10

1. Data Preparation

utt0000.A.1     0.37    1.76    A       bck     0       3       0       hmm hmm hmm .
utt0001.B.2     10.99   12.13   B       fra     1       2       0       are we
utt0002.B.3     12.13   14.53   B       el.inf  4       19      0       we're not allowed to dim the lights so people can see that a bit better ?
  • use data/meeting.py to convert the processed files into pickle binaries

2. Model Training

python train_ami.py
  • set the hyperparameters in args['...']
  • choose GPU(s) by setting os.environ["CUDA_VISIBLE_DEVICES"] = '0'

3. Model Decoding

python decode_ami.py start_idx
  • set the decoding parameters in args['...']

4. Model Evaluation (ROUGE summarisation)

PyROUGE: https://pypi.org/project/pyrouge/

5. Model Evaluation (multitask learning tasks)

python evaluate_attn_div_ami.py start_idx end_idx