question about reproducing on CIFAR100

[zhannn989]

Hello, Thanks for the awesome work! I am trying to reproduce your results but I have some questions about reproducing on CIFAR-100.

1. When I reproduce your results given your code, the results on CIFAR-10 is almost same with the results in paper. However, there is the huge gap about the results on CIFAR-100 with buffer 500. Following the example python main.py --dataset seq-cifar100 --model scommer --buffer_size 500 --experiment_id scommer-c100-500 --load_best_args you provide, I just get average accuracy about 40% on ema model while the results in your paper is 49%. Do I miss some details and do something wrong?

2. The parameters you choose for CLS-ER on CIFAR-100 are not in the best_args.py. Can you provide the parameters and the enviroment requirement documents as I find that the results of baselines in the paper are also higher?

Thanks for your answers in advance.