Introduction

Pytorch implementation from NUS DSML DSA5204 Group 12 for

• VPU (A Variational Approach for Learning from Positive and Unlabeled Data)
• nnPU (Positive-Unlabeled Learning with Non-Negative Risk Estimator)
• baseline model

Method to run

1. install conda environment with yaml file or with following command:
   • conda create -n DSA5204 python=3.8 pytorch=2.0.0 torchvision numpy scikit-learn -c pytorch
2. conda activate DSA5204
3. python run.py --dataset pageblocks --batch_size 200 --method vpu --gpu 0

Available options:

• --dataset {fashionMNIST, pageblocks}
• --batch_size
• --method {vpu, nnpu, raw}
• --gpu (Valid only when running on cuda available machines)
• --data_path
• --prior (*Only used for nnPU method)