HRN

Code for NeurIPS paper: "HRN: A Holistic Approach to One Class Learning"

Prerequisites

Some important packages' versions are as follow:
scikit-learn == 0.21.3
torch == 1.2.0

Usage

You can run our code on MNIST directly by this instruction "python3 main.py". Meaning of the arguments:

--max_epochs: the number of epochs of training

--batch_size: the size of the batches

--lr: the learning rate of the adam optimizer

--n_cpu: the number of cpu threads to use during batch generation

--img_size: the lenth of input image vectors (eg. mnist is 28*28=784)

--num_classes: the number of classes of the dataset

--gpu: choose whether to use gpu

--dataset: choose dataset for experiments

Please cite our paper the code helps you, thanks very much.
@article{hu2020hrn, title={HRN: A Holistic Approach to One Class Learning}, author={Hu, Wenpeng and Wang, Mengyu and Qin, Qi and Ma, Jinwen and Liu, Bing}, journal={Advances in Neural Information Processing Systems}, volume={33}, year={2020} }