Pytorch implementation of R-net

Model description

Model for reading comprehension style question answering based on R-net paper

Requirements

• Python 3.5 or higher
• PyTorch 1.0.1 or higher
• Numpy 1.16.2 or higher
• Spacy 2.1.2 or higher
• tensorboardX

Usage

Preproccessing

First of all, it's neccessary to create folder with dataset and glove embeddings in it, e.g. ./dataset/train-v1.1.json, ./dataset/dev-v1.1.json and ./dataset/glove.840B.300d.txt. Preprocessor builds vocabulary and converts text into tensors.

For preproccess use:

python preproccess.py

Training

For train model use:

python train.py

After training model is saving into ./models_dumps/ directory.

You can tune model with lots arguments available in model. Default configuration is used, but you can change it in ./config/train_config.yml.

Testing

For testing use:

python train.py --mode=test

See help of each module for more information and available arguments.