This is the official code of TransRate.

We provide a Jupyter Notebook(Demo.ipynb) for the demonstration of TransRate.

The calculation of TranRate requires $logdet$ on the data covariance matrix, which can be calculated using the eigenvalues of the data matrix. (see Lemma D.2). For demonstration convenience, we extract the eigenvalues and store them in ./logs_trans. The code for extracting the eigenvalues can be found in ./generate_transrate.

If you find this code useful for your research, please cite our paper:

@inproceedings{huang2022frustratingly,
  title={Frustratingly easy transferability estimation},
  author={Huang, Long-Kai and Huang, Junzhou and Rong, Yu and Yang, Qiang and Wei, Ying},
  booktitle={International Conference on Machine Learning},
  pages={9201--9225},
  year={2022},
  organization={PMLR}
}