This is a PyTorch implementation of Variational PU Learning


To train the model and reproduce the results in Table 1 and Table 2 in the original paper, each dataset with the first setting:

python train.py --dataset cifar10 --lam 0.03 --num_labeled 3000 --learning_rate 3e-5 --gpu <gpu_id>

python train.py --dataset fashionMNIST --lam 0.3 --num_labeled 3000 --learning-rate 3e-4 --gpu <gpu_id>

python train.py --dataset stl10 --lam 0.1 --num_labeled 2250 --learning-rate 1e-4 --gpu <gpu_id>

python train.py --dataset pageblocks --lam 0.0001 --num_labeled 100 --learning-rate 3e-4 --batch-size 100 --gpu <gpu_id>

python train.py --dataset grid --lam 0.1 --num_labeled 1000 --learning-rate 3e-4 --gpu <gpu_id>

python train.py --dataset avila --lam 0.1 --num_labeled 2000 --learning-rate 1e-4 --gpu <gpu_id>

