KiL
The source code of "Improving Text Classification with Knowledge in Labels" is provided here.
PS: The model is trained on a Tesla V100 32GB gpu. For the reproducibility, please use the same device.
Usage
1. Environment settings
1. First install libraries
$ pip install -r requirements.txt
2. download glove embeddings
please download pretrained embeddings from here. Select glove.6B.zip and extract glove.6B.300d.txt here.
2. Training LSTM model
You can choose dataset from {agnews, imdb, newsgroup} and select your gpu id.
$ python kil_lstm.py -d newsgroup -g 0
3. Training BERT model
You can choose dataset from {agnews, imdb, newsgroup} and select your gpu id.
$ CUDA_VISIBLE_DEVICES=0 python kil_bert.py -d newsgroup