vanderschaarlab/goggle

GOGGLE: Generative Modelling of Tabular Data By Learning Relational Structure

PyTorch implementation of ICLR'23 paper GOGGLE: Generative Modelling for Tabular Data by Learning Relational Structure. Authors: Tennison Liu, Zhaozhi Qian, Jeroen Berrevoets, Mihaela van der Schaar

---

Abstract

Generative modelling of tabular data entails a particular set of challenges, including heterogeneous relationships, limited number of samples, and difficulties in incorporating prior knowledge. This work introduces GOGGLE, a generative model that learns a relational structure underlying tabular data to better model variable dependencies, to introduce regularization, and to incorporate prior knowledge.

Key components of GOGGLE Framework. 1. Simultaneous learning of relational structure $G_\phi$ and $F_\theta$ s.t. generative process respects relational structure. 2. Injection of prior knowledge and regularization on variable dependence. 3. Synthetic sample generated using $\hat{x} = F_\theta(z; G_\phi) :, : z\sim p_Z$.

---

Experiments

To setup the virtual environment and necessary packages, please run the following commands:

$ conda create --name goggle_env python=3.8
$ conda activate goggle_env

Clone this repository and navigate to the root directory:

$ git clone https://github.com/tennisonliu/goggle.git
$ cd goggle

Install the required modules:

$ pip install -r requirements.txt

Place dataset in exps/data, and see experiment notebooks with instructions in:

• 1. exps/synthetic_data. for synthetic data generation
• 2. exps/prior_knowledge/. for incorporating prior knowledge, and
• 3. exps/ablation/. for ablation settings.

---

Citation

If our paper or code helped you in your own research, please cite our work as:

@inproceedings{liu2023goggle,
  title={GOGGLE: Generative Modelling for Tabular Data by Learning Relational Structure},
  author={Liu, Tennison and Qian, Zhaozhi and Berrevoets, Jeroen and van der Schaar, Mihaela},
  booktitle={International Conference on Learning Representations},
  year={2023}
}