Project Details emphasized in the paper.
-
python main_Script.py main_script.py contains the Bert model and the training part. After the execution a directory named Bert would be created which will contain the predictions file
-
python evaluate-v1.0 --data-file data/coqa-dev-v1.0.json --pred-file Bert/predictions.json After the training is complete and the predictions are made, run the evaluate-v1.0 file (downloaded from https://stanfordnlp.github.io/coqa/) with the above arguments to get the results.
Files: Main_script.py => Contains the Bert model and the training part. evaluate-v1.0 => official CoQA evaluation script Data => coqa-dev-v1.0.json (Development dataset) => coqa-train-v1.0 (Training dataset) =>processors => metrics.py (For computing the predictions from the output logits) => coqa.py (Pre-processing) => utils.py (Utility file that supplements pre-processing)