Detailed setup to reproduce Table 1

Question

Detailed setup to reproduce Table 1

KAI-YUE opened this issue 2 years ago · 2 comments

Is it possible to provide detailed setup to reproduce Table 1 in the paper. I have tried multiple combinations of parameters, but the coreset results are not comparable to random purning. Besides, the accuracy is not as high as Table 1 reported.

Answer 1 · 2023-04-24T15:18:39.000Z

For Tab. 1, here is the parameters.
python -u main.py --fraction 0.5 --dataset CIFAR10 --data_path ~/datasets --num_exp 1 --workers 10 --optimizer SGD -se 10 --selection Glister --model ResNet18 --lr 0.1 -sp ./result --batch 128 --scheduler CosineAnnealingLR

Answer 2 · 2024-05-14T22:45:15.000Z

Hi @Chengcheng-Guo thank you very much. As a follow up to this, how are the standard deviation values reported? i.e do you run this script for a certain number of times and then report the results? or how many random seeds are used?
thanks!