Our final, best performing model is the bi-encoder model for evidence retrieval and a sequential model for claim classification named Tue9am_Group6.ipynb. The log output for eval.py is outputted within the model's jupyter notebook itself.
Our transformer model consists of two parts, the evidence retrieval (Transformer_Evidence_Retrieval.ipynb) file and the label classification (Transformer_Label_Classification.ipynb) file. The evidence retrieval file has to be run to create the file for label classification. Label classification also has commented out code to use the bi-directional LSTM model's retrieved evidences (in the file, these have comments relating to using Noah's retrieved evidences).
Our bi-directional LSTM model file for evidence retreival is called Bi_Directional_LSTM.ipynb.
- Install required packages before running the notebooks (keras, NLTK etc)
- Maybe need to change the path to the data files before opening the files