https://github.com/rz-zhang/SeqMix

The repository for our EMNLP'20 paper SeqMix: Augmenting Active Sequence Labeling via Sequence Mixup.
https://github.com/rz-zhang/SeqMix

Last synced: 3 months ago
JSON representation

The repository for our EMNLP'20 paper SeqMix: Augmenting Active Sequence Labeling via Sequence Mixup.

• Host: GitHub
• URL: https://github.com/rz-zhang/SeqMix
• Owner: rz-zhang
• Created: 2020-09-28T09:42:26.000Z (about 4 years ago)
• Default Branch: master
• Last Pushed: 2021-09-05T07:20:20.000Z (about 3 years ago)
• Last Synced: 2024-06-24T05:47:54.317Z (5 months ago)
• Language: Python
• Size: 1.85 MB
• Stars: 42
• Watchers: 3
• Forks: 6
• Open Issues: 3
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

• StarryDivineSky - rz-zhang/SeqMix
• Awesome-Mixup - [Code

README

        
# SeqMix

The repository of our EMNLP'20 paper  

**SeqMix: Augmenting Active Sequence Labeling via Sequence Mixup**  

[[paper]](https://rongzhizhang.org/pdf/emnlp20_SeqMix.pdf)  [[slides]](https://rongzhizhang.org/slides/EMNLP20_SeqMix_Slides.pdf)  

  

![Illustration of the three variants of SeqMix](SeqMix.png)

# Requirements

- pytorch-transformers==1.2.0  

- torch==1.2.0  

- seqeval==0.0.5  

- tqdm==4.31.1  

- nltk==3.4.5  

- Flask==1.1.1  

- Flask-Cors==3.0.8  

- pytorch_pretrained_bert==0.6.2  

Install the required packages:  

```

pip install -r requirements.txt

```

# Key Parameters

- `data_dir`: specify the data file, we provide CoNLL-03 dataset here 

- `max_seq_length`: maximum length of each sequence  

- `num_train_epochs`: number of training epochs  

- `train_batch_size`: batch size during model training  

- `active_policy`: query policy of active learning  

- `augment_method`: augmenting method  

- `augment_rate`: augmenting rate  

- `hyper_alpha`: parameter of Beta distribution

# Run

## Active learning part

Random Sampling  

```

python active_learn.py --active_policy=random

```

Least Confidence Sampling  

```

python active_learn.py --active_policy=lc

```

Normalized Token Entropy sampling  

```

python active_learn.py --active_policy=nte

```

## Seqmix part

Whole sequence mixup

```

python active_learn.py --augment_method=soft

```

Sub-sequence mixup

```

python active_learn.py --augment_method=slack

```

Label-constrained sub-sequence mixup

```

python active_learn.py --augment_method=lf

```