/CBA-MR-noisy-label

Primary LanguagePythonOtherNOASSERTION

When False Positive is Intolerant

An implement of the NeurIPS 2021 paper: When False Positive is Intolerant: End-to-End Optimization with Low FPR for Multipartite Ranking

Environments

  • Ubuntu 16.04
  • CUDA 11.1
  • Python 3.8.0
  • Pytorch 1.7.0

See requirement.txt.

Data preparation

Download datasets from Google drive and unzip it to ./data.

Training

  1. Modify configs in scripts/[dataset]/train_cba.sh
  2. Run the script:
sh `scripts/[dataset]/train_cba.sh`

The model and log are saved in output/[dataset]/logit_cba by default.

Evaluation

  1. Download the pretrained model from Google drive.
  2. Modify configures in scripts/[dataset]/eval_cba.yaml: change --checkpoint to the path where the model is saved.
  3. Run
sh `scripts/[dataset]/eval_cba.sh`

The results might slightly differ from the above due to the environment difference in the training process.

References

If this code is helpful to you, please consider citing our paper:

@inproceedings{wen2021false,
  title={When False Positive is Intolerant: End-to-End Optimization with Low FPR for Multipartite Ranking},
  author={Wen, Peisong and Xu, Qianqian and Yang, Zhiyong and He, Yuan and Huang, Qingming},
  booktitle={Thirty-Fifth Conference on Neural Information Processing Systems},
  year={2021}
}