The mainly purpose of starting this project is to build simple & readable deep learning neural relation extraction tutorials for beginners who want to study natural language processing.
This project is organized clearly.
-
Models are stored in ./models folder. You can still find descriptions of models in ./models folder.
-
Dataset contains all various common RE dataset. - E.g.
dataset/sem-task8
is one of RE dataset, and all original dataset files are contained indataset/sem-task8
folder directly. Indataset/sem-task8
you can still findsdataset/sem-task8/train
,dataset/sem-task8/test
, they are generated bydataset/semProcessor.py
to build the SEMData, a subclass of torch.utils.data.Dataset, which used to build dataLoader. -
Checkpoints : used to record your best model parameters, which is now a empty folder.
-
source : store all images used in README.md
-
semeval : contains evaluate func which is provide by official dataset owner, details are showed in
semeval/README.md
. -
config.py : to test all provide model in this project, all you need is to change this file. contains which model you want to train & all parameters you can tunes.
-
main.py : contains iterations on train & test dataset, and generate result.txt into
semeval
dataset.