Reproducing CIFAR10 experiments

[j0807s]

First of all, thank you for sharing the code for the community!

I am trying to reproduce Table 1 but I got the lower accuracy than reported results on CIFAR10 dataset.

I got the results of 10 runs below:

Memory 100 200 500 1000 33.76±1.6 40.03±1.67 46.43±1.89 48.13±2.7

Can you check if it is right?

Thanks.

[FelixHuiweiLin]

First of all, thank you for sharing the code for the community!

I am trying to reproduce Table 1 but I got the lower accuracy than reported results on CIFAR10 dataset.

I got the results of 10 runs below:

Memory 100 200 500 1000 33.76±1.6 40.03±1.67 46.43±1.89 48.13±2.7

Can you check if it is right?

Thanks.

Hi,

I confirm that the code is OK. I think the result you mentioned is the result of dividing CIFAR10 into 10 tasks. However, CIFAR10 is divided into 5 tasks in our paper. Please confirm whether it is true.

Thanks.

[comewei]

hello, when i use --num_runs 1 --data cifar10 --cl_type nc --agent PCR --retrieve random --update random --mem_size 200 ,when it runs, it show the message RuntimeWarning: Degrees of freedom <= 0 for slice keepdims=keepdims) in the avg_end_acc, avg_end_fgt, avg_acc, avg_bwtp, avg_fwt = compute_performance(accuracy_array) print('----------- Total {} run: {}s -----------'.format(params.num_runs, end - start)) print('----------- Avg_End_Acc {} Avg_End_Fgt {} Avg_Acc {} Avg_Bwtp {} Avg_Fwt {}-----------' .format(avg_end_acc, avg_end_fgt, avg_acc, avg_bwtp, avg_fwt)) in the run.py