AnatoliiPotapov/squad

This is not yet complete!

Experiments on SQuAD dataset with Keras. The purpose is to make clean and efficient architecture which is easily reproducible and achieve close to state of the art results.

This work is highly inspired by DRQA here and FastQA there

Results

Instructions

1. We need to parse and split the data

    python parse_data.py data/train-v1.1.json --train_ratio 0.9 --outfile data/train_parsed.json --outfile_valid data/valid_parsed.json
    python parse_data.py data/train-v1.1.json --outfile data/train_parsed.json

2. Preprocess the data

    python preprocessing.py data/train_parsed.json --outfile data/train_data.pkl
    python preprocessing.py data/valid_parsed.json --outfile data/valid_data.pkl
    python preprocessing.py data/dev_parsed.json --outfile data/dev_data.pkl

3. Train the model

    python train.py --hdim 40 --batch_size 70 --nb_epochs 50 --optimizer adam --dropout 0.2

4. Predict on dev/test set samples

    python predict.py model/your-model prediction.json