Code for DL4NLP project by Paras Dahal, Koen Gommers, Kevin Waller, Danai Xezonaki. Using custom seq2seq model for the Text2SQL task. Baseline model consists of BERT as an encoder and a LSTM with attention as a decoder. We extend our baseline model by concatenating schema information to the question input.
Download Spider dataset and put the data in a datasets/spider directory.
Install required packages
pip install -r requirements.txt
Training (all arguments can be found in utils.py)
python train.py --seed=42 --use_schema=False
By default, output will be saved in the logs directory.
Output CSVs can be evaluated using the eval.py
python eval.py outputs.csv