Temporal code repo for "Mitigating poisoning attack by Shapley value"
Hard to read, many debug informations contained. But it can be executed well and flexible to modify its structure.
A lot of code-clean work to be done.. :(
First execute run.py with parameters eps = 0.2, #poisoning data = 0.2 * 360 = 240
python run.py --dataset "MNIST | CIFAR | ImageNet" --eps 0.2 --num 240 >> resultsnew.txt
Then run rubbish.py to collect data from "resultsnew.txt" to generate "poison.pkl"
python rubbish.py --num 240
Finally run eval.py to evaluate the performance made by Oracle, Shapley value, influence function and random removal:
python eval.py --dataset "MNIST | CIFAR | ImageNet" --eps 0.2 --num 240