Why even use PyTorch's DataLoader in dataloader.py

[LexSong]

It seems that the only way to get data from self.dataloader, is from get_batch(), and in get_batch() we recreate a new iterator of self.dataloader every time it is called. Therefore we always use only the first item returned by self.dataloader, which looks very pointless to create self.dataloader in the first place.

[akanimax]

Hi @LexSong,

You are right about the dataloader always returning first element of the iterator. But since the dataloader while creation has set shuffle=True, It will return different data-points (well, hopefully 😄). I wrote a package for pytorch for ProGANs (well structured code), you can check it out at https://github.com/akanimax/pro_gan_pytorch. Examples for using this package are at https://github.com/akanimax/pro_gan_pytorch-examples which validates that my code works.

Cheers!

[LexSong]

It's really inefficient though.

By default RandomSampler chooses samples without replacements, so it will generate a random indices list (by torch.randperm(n).tolist()) every time we call it. If randomly sampling with replacements is what we want here, just create a dataset properly.

[akanimax]

@LexSong, Yes, I know it is inefficient. BTW, I am not the author of this code, was just helping. Did you check out the links?

[shamanez]

@akanimax your codes are perfect. I even followed the VDB code which is very efficient and informative.