BiDAF_pytorch

Implementing the Bidirectional Attention Flow model using pytorch

0. configure your environment:

    - create a conda env : conda create {"env name"} python= 3.6
    - activate {"env name"}
    - install all requirements 
   

1. preprocessing your dataset :

    - run  python setup.py
            * this will preprocess SQuAD 2.0 traing , dev , test and my_test datasets ( dataset containing a concrete example) 
            * this will also dowlnlods the Glove 300-dimensional word vectors
   

2. train your model :

    - create "save" folder
    - run python train.py --name training 
            * this will train your model by reading command line arguments, loading SQuAD dataset
            * you can add --num_epochs argument to specify the number of epochs 
            * with 10 epochs you will have : F1: 58.47 , EM: 55.42 , AvNA: 64.88
   

3. to test your model with the dev/test SQUAD dataset:

    - run python test.py --name dev/test --split dev/test --load_path ./save/{"your training folder"}/best.pth.tar
   

4. to visualize the machine learning workflow in tensorboard :

    - run tensorboard --logdir ./save/{"your training folder"}
   

5. to test your model with your test dataset

    - run python my_test.py -name test --split test  --load_path ./save/{"your training folder"}/best.pth.tar