RecTable: Fast Modeling Tabular Data with Rectified Flow

This repository contains the implementation of RecTable: Fast Modeling Tabular Data with Rectified Flow.

Installation

The environments of our experiments are based on PyTorch 2.3.1 (docker image)

Pull docker image.

docker pull pytorch/pytorch:2.3.1-cuda11.8-cudnn8-devel

Install other packages in the container.

pip install -r requirements.txt

Preparing Datasets

Using the datasets experimented in the paper

Download raw dataset:

python download_dataset.py

Process dataset:

python process_dataset.py

First, create a directory for you dataset [NAME_OF_DATASET] in ./data:

cd data
mkdir [NAME_OF_DATASET]

Put the tabular data in .csv format in this directory ([NAME_OF_DATASET].csv). The first row should be the header indicating the name of each column, and the remaining rows are records.

Then, write a .json file ([NAME_OF_DATASET].json) recording the metadata of the tabular, covering the following information:

{
    "name": "[NAME_OF_DATASET]",
    "task_type": "[NAME_OF_TASK]", # binclass or regression
    "header": "infer",
    "column_names": null,
    "num_col_idx": [LIST],  # list of indices of numerical columns
    "cat_col_idx": [LIST],  # list of indices of categorical columns
    "target_col_idx": [list], # list of indices of the target columns (for MLE)
    "file_type": "csv",
    "data_path": "data/[NAME_OF_DATASET]/[NAME_OF_DATASET].csv"
    "test_path": null,
}

Put this .json file in the .Info directory.

Finally, run the following command to process the UDF dataset:

python process_dataset.py --dataname [NAME_OF_DATASET]

Training

python main.py --dataname [NAME_OF_DATASET] --exp "rectable_[NAME_OF_DATASET]"

Other configurations can be found in config.py.

Generation

python generate.py --ckpt CKPT_PATH --dataname [NAME_OF_DATASET] 

CKPT_PATH is same of exp in the configuration when training. For example, when the checkpoint's path is checkpoints/rectable_sample.pth, the generation command as follows:

python generate.py --ckpt sample --dataname [NAME_OF_DATASET]

Acknowledgement

This repo is built upon the previous work TabSyn's, TabDiff's, and Rectified Flow. Many thanks to Hengrui, Shi, and Liu!

Citation

We appreciate your citations if you find this repository useful to your research!

@misc{fuchi2025rectablefastmodelingtabular,
      title={RecTable: Fast Modeling Tabular Data with Rectified Flow}, 
      author={Masane Fuchi and Tomohiro Takagi},
      year={2025},
      eprint={2503.20731},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2503.20731}, 
}

Contact

The email in the paper is availabe until the end of March, 2025 due to author's graduation. Please contact us via Twitter (X). Twitter account can be found my profile page.