- Install requirements:
pip install -r requirements.txt
- Download BERT base model (uncased) and place it in
/bert
- To run train.py:
python train.py
- To run train.py in debug mode:
python train.py --debug True
- To run the best saved model, download the checkpoint, unzip it, and place it in
/saved_models
. - Then, run
python eval.py
.
- Logs for any given run are kept in
/log
- The preprocessing script in
/preprocessing
combines the raw wine data and splits it randomly into train/dev/test splits (80/10/10).