ncNet

Supporting the translation from natural language (NL) query to visualization (NL2VIS) can simplify the creation of data visualizations because if successful, anyone can generate visualizations by their natural language from the tabular data.

We present ncNet, a Transformer-based model for supporting NL2VIS, with several novel visualization-aware optimizations, including using attention-forcing to optimize the learning process, and visualization-aware rendering to produce better visualization results.

Please refer to our paper at IEEE VIS 2021 for more details.

Environment Setup

Python3.6+
PyTorch 1.7
torchtext 0.8
ipyvega

Install Python dependency via pip install -r requirements.txt when the environment of Python and Pytorch is setup.

Running Code

Data preparation

[Must] Download the Spider data here and unzip under ./dataset/ directory
[Optional] Only if you change the train/dev/test.csv under the ./dataset/ folder, you need to run process_dataset.py under the preprocessing foler.

Runing Example

Open the ncNet.ipynb to try the running example.

Training

Run train.py to train ncNet.

Testing

Run test.py to eval ncNet.

Citing ncNet

@inproceedings{ncnet,
author    = {Yuyu Luo and Nan Tang and Guoliang Li and Jiawei Tang and Chengliang Chai and Xuedi Qin},
title     = {Natural Language to Visualization by Neural Machine Translation},
booktitle = {{IEEE Visualization Conference (VIS)}},
year      = {2021},
}

License

The project is available under the MIT License.

Contact

If you have any questions, feel free contact Yuyu Luo (luoyy18 [AT] mails.tsinghua.edu.cn).

guoyi118/ncNet-1