/ABL_nc

Primary LanguagePython

🌟 New! ABLkit released: A toolkit for Abductive Learning with high flexibility, user-friendly interface, and optimized performance. Welcome to try it out!🚀

Enabling Knowledge Refinement upon New Concepts in Abductive Learning

This is the repository for holding the sample code of Enabling Knowledge Refinement upon New Concepts in Abductive Learning in AAAI 2023.

This code is only tested in Linux environment.

Environment Dependency

To create the above environment with Anaconda, you can run the following command (cudatoolkit=11.3 for new GPUs / new drivers, cudatoolkit=10.1 for old GPUs):

(cudatoolkit=11.3)

conda create -n ablnc python=3.8 -y
conda activate ablnc
conda install pytorch=1.12 torchvision torchaudio cudatoolkit=11.3 -c pytorch
pip install cupy-cuda11x clingo tqdm matplotlib imblearn pytod scikit-learn
Download and install ILASP according to https://doc.ilasp.com/installation.html and copy './ILASP' to current path

(cudatoolkit=10.1)

conda create -n ablnc python=3.8 -y
conda activate ablnc
conda install pytorch=1.7 torchvision torchaudio cudatoolkit=10.1 -c pytorch
pip install cupy-cuda101 clingo tqdm matplotlib imblearn pytod scikit-learn
Download and install ILASP according to https://doc.ilasp.com/installation.html and copy './ILASP' to current path

Running Code

To reproduce the experiment results, we can simply run the following code:

  • Less-Than with New Digits

    python main.py --task=less_than

  • Chess with New Pieces

    python main.py --task=chess

  • Multiples of Three

    python main.py --task=multiples_of_three

To view or change the hyperparameters, please refer to the arg_init() function in the code.

Reference

@inproceedings{ablnc2023huang,
  title={Enabling Knowledge Refinement upon New Concepts in Abuctive Learning},
  author={Huang, Yu-Xuan and Dai, Wang-Zhou and Jiang, Yuan and Zhou, Zhi-Hua},
  booktitle={Proceedings of the 37th AAAI Conference on Artificial Intelligence (AAAI'23)},
  //pages={},
  year={2023}
}