Points for sampler

[halbielee]

Thanks for sharing your great job.

I have a question about the point for sampler in your code.

train_sampler is first declared at L167 in cifar_train.py then, train_loader gets the sampler in L169-171.

This seems to be fine in itself. But in the middle of training(train+validation) there is a part which seems to be for sampler in L186-L208. I understand this part is need for LDAM and DRW, but I think this new train_sampler object does not affect train_loader.

How's your opinion?
Thnks!

LDAM-DRW/cifar_train.py

Lines 186 to 208 in 3193f05

	if args.train_rule == 'None':
	train_sampler = None
	per_cls_weights = None
	elif args.train_rule == 'Resample':
	train_sampler = ImbalancedDatasetSampler(train_dataset)
	per_cls_weights = None
	elif args.train_rule == 'Reweight':
	train_sampler = None
	beta = 0.9999
	effective_num = 1.0 - np.power(beta, cls_num_list)
	per_cls_weights = (1.0 - beta) / np.array(effective_num)
	per_cls_weights = per_cls_weights / np.sum(per_cls_weights) * len(cls_num_list)
	per_cls_weights = torch.FloatTensor(per_cls_weights).cuda(args.gpu)
	elif args.train_rule == 'DRW':
	train_sampler = None
	idx = epoch // 160
	betas = [0, 0.9999]
	effective_num = 1.0 - np.power(betas[idx], cls_num_list)
	per_cls_weights = (1.0 - betas[idx]) / np.array(effective_num)
	per_cls_weights = per_cls_weights / np.sum(per_cls_weights) * len(cls_num_list)
	per_cls_weights = torch.FloatTensor(per_cls_weights).cuda(args.gpu)
	else:
	warnings.warn('Sample rule is not listed')

[AlanChou]

I think you're right. I also found this issue when running the code. I think it only effects "Resample" and can be fixed by adding a line after the "ImbalancedDatasetSampler" is created to make sure the dataloader use the created sampler.

[halbielee]

Thank you for the quick response!

[AlanChou]

You should probably wait for the author's response before closing the issue though I think you're right about the question. And it's also better to keep the issue opened before the code is updated.
I’m just a random guy which want to help :)

[halbielee]

@AlanChou
Oh, I see. I thought you were the author. :)

Thank you for the another reply!

[yibuxulong]

The same question for me. I found it rejected to change the sampler of train_loader after initialization(because the sampler affects shuffle, etc.), so I wonder if I should:

1. resample and create train_loader each epoch
2. just resample and create train_loader once before the epoch-loop