forked from Rahul-Vigneswaran K's repo.
Here's an example argument set.
--lr=2e-4
--batch_size=60
--end_iter=25
--resume
--gpu=1
--dataset=cifar10
--arch_type=conv2
--prune_percent=10
--prune_iterations=27
--split_conv_and_fc
--fc_prune_percent=20
For now there's a weird bug that makes every #param-testacc graph identical kinda same in terms of shape when trying experiments multiple time.
I think it could be a saving/loading mechanism from original repo, i should fix asap. (or test with other params?)