Skip to content

Latest commit

 

History

History
60 lines (42 loc) · 2.23 KB

README.md

File metadata and controls

60 lines (42 loc) · 2.23 KB

Turning Tabular Foundation Models into Graph Foundation Models

This is the official repository for the paper "Turning Tabular Foundation Models into Graph Foundation Models" (arXiv). In this repository, we provide code for reproducing our experiments with G2T-FM, GNNs and LightGBM (including ablation). Code for reproduction of our experiments with prior GFMs is coming soon.

Note

Our work is largely based on TabPFN, LimiX and "On Finetuning Tabular Foundation Models" paper, please consider checking them out too!

Reproducing Experiments

Prerequisites

  1. Install uv
  2. Install dependencies
uv sync
  1. Download TabPFNv2 checkpoints
wget https://huggingface.co/Prior-Labs/TabPFN-v2-reg/resolve/main/tabpfn-v2-regressor.ckpt?download=true -O checkpoints/tabpfn-v2-regressor.ckpt
wget https://huggingface.co/Prior-Labs/TabPFN-v2-clf/resolve/main/tabpfn-v2-classifier.ckpt?download=true -O checkpoints/tabpfn-v2-classifier.ckpt
  1. For experiments on GraphLand, download datasets and place them in "data" directory

Running the code

You can execute a minimal run with a following command:

uv run bin/go.py exp/g2t_tabpfnv2/finetune/tolokers-2/tuning.toml --force

Project Structure

  • bin/ - Training and evaluation scripts
  • exp/ - Experiment configurations and results
  • data/ - Dataset directory (created after download)
  • lib/ - Common utilities and tools

Configuration

Experiments are configured using TOML files located in the exp/ directory. Each configuration specifies:

  • Dataset path and preprocessing
  • Model hyperparameters
  • Training settings
  • Evaluation metrics

Results

After training, results are saved in the same directory as the configuration file:

  • report.json - Evaluation metrics
  • Model checkpoints
  • Training logs

Licenses

This project includes third-party components (TabPFN and LimiX) under separate licenses. See LICENSES/ and NOTICE for details.

Built with PriorLabs-TabPFN