takahashihiroshi/puae

pytorch implementation for "Deep Positive-Unlabeled Anomaly Detection for Contaminated Unlabeled Data".

Deep Positive-Unlabeled Anomaly Detection for Contaminated Unlabeled Data

This is a pytorch implementation of the following paper [arXiv]:

@misc{takahashi2024deep,
      title={Deep Positive-Unlabeled Anomaly Detection for Contaminated Unlabeled Data}, 
      author={Hiroshi Takahashi and Tomoharu Iwata and Atsutoshi Kumagai and Yuuki Yamanaka},
      year={2024},
      eprint={2405.18929},
      archivePrefix={arXiv},
      primaryClass={stat.ML}
}

Please read LICENCE.md before reading or using the files.

Prerequisites

• Please install python>=3.10, numpy, scipy, torch, torchvision, scikit_learn, and matplotlib
• Please also see requirements.txt
• Note that we use the nvidia pytorch docker image (23.12-py3)
  • https://catalog.ngc.nvidia.com/orgs/nvidia/containers/pytorch/tags

Datasets

All datasets will be downloaded when first used.

Usage

for image datasets

usage: main.py [-h] [--dataset DATASET] [--normal_class NORMAL_CLASS]
               [--unseen_anomaly UNSEEN_ANOMALY] [--algorithm ALGORITHM]
               [--alpha ALPHA] [--n_epoch N_EPOCH]
               [--learning_rate LEARNING_RATE] [--batch_size BATCH_SIZE]
               [--seed SEED]

• You can choose the dataset from following datasets:
  • MNIST: handwritten digits
  • FashionMNIST: fashion product images
  • SVHN: house number digits
  • CIFAR10: animal and vehicle images
• You can choose the normal_class from 0 to 9
• You can choose the unseen_anomaly from 0 to 9
• You can choose the algorithm from following algorithms:
  • AE: Autoencoder
  • DeepSVDD: DeepSVDD
  • ABC: Autoencoding Binary Classifier
  • DeepSAD: Deep Semi-Supervised Anomaly Detection
  • PU: PU Learning Classifier
  • PUAE: Our approach with AE
  • PUSVDD: Our approach with DeepSVDD
• You can change the alpha, the hyperparameter of PU learning-based approaches
• You can change the random seed of the training and n_epoch, learning_rate, and batch_size of the optimizer

for toy dataset

usage: toy.py [-h] [--algorithm ALGORITHM] [--alpha ALPHA] [--n_epoch N_EPOCH]
              [--learning_rate LEARNING_RATE] [--batch_size BATCH_SIZE]
              [--seed SEED]

• You can choose the algorithm from following algorithms:
  • AE: Autoencoder
  • DeepSVDD: DeepSVDD
  • ABC: Autoencoding Binary Classifier
  • DeepSAD: Deep Semi-Supervised Anomaly Detection
  • PU: PU Learning Classifier
  • PUAE: Our approach with AE
  • PUSVDD: Our approach with DeepSVDD
• You can change the alpha, the hyperparameter of PU learning-based approaches
• You can change the random seed of the training and n_epoch, learning_rate, and batch_size of the optimizer

Example

MNIST experiment (normal: 1 / unseen: 0) with our approach:

python main.py --dataset MNIST --normal_class 1 --unseen_anomaly 0 --algorithm PUAE