A brief outline of important files:
- permuteParsWithId.ipynb contains code to generate a dataset given an arbitrary corpus of texts
- The 'Simple GRU' Network along with relevant dataset and training functions is written in coherenceModel.py
- The training script for 'Simple GRU' networks is trainAvStyle.py
- The 'Bidirectional GRU' Network along with relevant dataset and training functionsis written in coherenceModelNews.py
- The training script for 'Bidirectional GRU' networks is trainLAStyle.py
- The script to use a trained model to score one dataset generated by permuteParsWithId.ipynb is scoreData.py (with a script to score all data in scoreAllData.py)
- Final models of various training sessions are all named best_rnn_*.pt with specific training corpuses denoted in asterisk
All of our datasets and figures we generated can be found in our google drive (visible to anyone at UofM): https://drive.google.com/drive/folders/1lcvrGD5Lp4L2MMZDefV3ZRPh1HVZN-R-?usp=sharing