Back in the days Torch
had a nice collection of data loaders. In sequence modeling for getting the [25 x 32 x 100]
tensor for RNN training I used SequenceLoader
which accepted (sequence, batchsize, bidirectional)
parameters where 25
is a sequence length, 32
is the batch size and 100
is the embedding size. Quickly skimming through current torch.utils.data
docs I found nothing for sequence, maybe I am not seeing it but how do you do sequence batching in pytorch now?