A QA System on SQuAD
First of all, you need to download dataset.
$ ./download.sh
You can use presaved pickle files or build data dictionaries with process_data.py
.
$ python train.py
$ python generate_predictions.py -m saved_keras_model_path -o prediction_file_name
Use an official script to obtain evaluation result.
$ python evaluate-v1.1.py ./dataset/dev-v1.1.json prediction_file_name
{"exact_match": XXXX, "f1": YYYY}