SSCL-Text2SQL

Code for our AAAI-23 accepted paper: Learn from Yesterday: A Semi-Supervised Continual Learning Method for Supervision-Limited Text-to-SQL Task Streams

Running Code

1. Preprocess data before training

We have provided IRNet readable form of data in data. You can also run our script to convert the raw task streams into this form.

cd preprocess
sh run_me_spider.sh
sh run_me_wikisql.sh

2. Training SFNet on task streams

Execute the following command for training on Spider task stream.

sh train_sfnet_spider.sh

Execute the following command for training on WikiSQL task stream.

sh train_sfnet_wikisql.sh

The model's loss and metrics are logged in ./saved_model/{your_run_id}/log. You can view them by running the following command to start TensorBoard.

tensorboard --logdir=./saved_model/{your_run_id}/log

Then, open your browser and enter localhost:6006 in the address bar to get to the TensorBoard page.