Neural Question Generation
This is the implementation of Neural Question Generation. The details can be found here
Requirements
Run experiments under Python 3.6.5 with following libs:
- tensorflow 0.12 with CUDA 8.0 support (optional)
- nltk 3.1
and data processing lib including
- spacy 1.8.1
- enchant 1.6.9
- joblib 0.11
Usage
Prepare
download Squad-v1.1 into ./data
download GloVe pretrained embeddings
checkout the default settings in ./preprocess/config.py
Data collection and dataset creation
run python config.py
Once finished, a random split of training/test dataset will be available in ./data
Train model
python run.pyRunning
python run.py --helpwill show a list of argument to configure the model.
You can press Ctrl-C anytime to stop training and start doing test on the test set with the best model evaluated on a randomly generated validation set from training set.
All the output will be available in ./log/{train,test}.log.