LoCalPFN

This is the codebase pertaining to the paper Retrieval & Fine-Tuning for In-Context Tabular Models, published in the main conference track of Advances in Neural Information Processing Systems 37 (NeurIPS 2024).

Preparation

Please download the TabZilla datasets by first cloning the repo

git clone https://github.com/naszilla/tabzilla

And download the datasets

cd tabzilla
python tabzilla_data_preprocessing.py --process_all

After that, please clone this repo and softlink the TabZilla dataset folder

git clone git@github.com:layer6ai-labs/LoCalPFN.git
cd LoCalPFN
mkdir datasets
ln -s datasets ~/tabzilla

Run Instructions

TabPFN-kNN

python main.py --exp_name="default_knn" knn

LoCalPFN

python main.py --exp_name="localpfn" ft

Citation

@inproceedings{thomas2024retrieval,
  title={Retrieval \& Fine-Tuning for In-Context Tabular Models},
  author={Thomas, Valentin and Ma, Junwei and Hosseinzadeh, Rasa and Golestan, Keyvan and Yu, Guangwei and Volkovs, Maksims and Caterini, Anthony},
  booktitle={Advances in Neural Information Processing Systems},
  year={2024}
}