ankitvad/SEASS-pytorch

Pytorch implementation for paper "Selective Encoding for Abstractive Sentence Summarization"

Pytorch implementation for paper "Selective Encoding for Abstractive Sentence Summarization"

• author: Kirk
• mail: cotitan@outlook.com

Requirments

• torch==0.4.0
• numpy==1.12.1+
• python=3.5+

Noticement

1. When running train.py, the console may print out EOFError or ConnectionResetError, which is caused by DataLoader module of pytorch. I don't know why but it does not influence the training process. Neglect It! Set num_workers=0 (argument of DataLoader) to avoid this error messages.
2. This project haven't finish yet, there are bugs to be fix, and modules to be implement.

How-to

1. Run python train.py to train
2. Run python mytest.py to generate summaries

TODO

1. Implement maxout layer
2. ROUGE metric
3. Attention calculation
4. Fix bugs on decoder