KeyError: 'train_subset'

[hkunzhe]

Hi, thanks for sharing your codes!
I was able to run perturbation.py and main.py in the Sample-wise noise for unlearnable example on CIFAR-10 section. However, when I try to run perturbation.py in the Class-wise noise for unlearnable example on CIFAR-10 section, it raises the following error:

Traceback (most recent call last):
  File "perturbation.py", line 483, in <module>
    main()
  File "perturbation.py", line 469, in main
    noise = universal_perturbation(noise_generator, trainer, evaluator, model, criterion, optimizer, scheduler, random_noise, ENV)
  File "perturbation.py", line 191, in universal_perturbation
    for i, (images, labels) in tqdm(enumerate(data_loader[args.universal_train_target]), total=len(data_loader[args.universal_train_target])):
KeyError: 'train_subset'

Thus, I add an option --universal_train_target train_dataset to fix this error. Is this right to get the class-wise perturbation?
BTW, there are two typos (--perturb_type samplewse=>--perturb_type samplewise) in README.md.

[HanxunH]

Hi,

Thanks for your interest in our work.

• Yes, --universal_train_target train_dataset should be added to the options for classwise setting . The README is missing it. In case of other errors, for each corresponding experiment in the paper, there is exp_setting.sh documents all the options.
• Thanks for the typos.
• I have updated the README for the --universal_train_target train_dataset option as well as fixed the typos.

Best,

[hkunzhe]

Thanks for your quick reply!
It seems you have added the --universal_train_target train_subset rather than --universal_train_target train_dataset in the README.md.

[HanxunH]

Sorry for the confusion.

I have double checked this. For classwise, as stated in the paper, we use We use 20% of the training dataset., so this should indeed be tran_subset. To avoid the error, please use --use_subset in perturbation.py

Technically, it is also ok to use entire dataset to generate the noise. We use 20% to simulates that it can generalize to unseen data (the rest of 80%).

[hkunzhe]

Thanks for your clarification!