Issues regarding datawise unlearning #5
Description
Hi @ljcc0930 @a-F1 :):)
the command-python main_selmu.py --arch resnet18 --dataset cifar10 --cp_path results/cifar10_resnet/origin/0model_SA_best.pth.tar --select_epochs 2 --num_indexes_to_replace 4500 --unlearn w_retrain --unlearn_steps 2 --theta_lr 1e-3 --save_dir results/cifar10_resnet/data-wise
followed by
python main_evalmu.py --arch resnet18 --dataset cifar10 --cp_path results/cifar10_resnet/origin/0model_SA_best.pth.tar --unlearn retrain --num_indexes_to_replace 4500 --unlearn_steps 182 --theta_lr 0.1 --w_path results/cifar10_resnet/data-wise/select_weight.pth.tar --save_dir results/cifar10_resnet/evaluation
yields UA almost same as RA I tried but i am unable to understand why is it not reducing to 5% as mentioned in the paper ?