@ptrblck_de I have a variable length data and I want to first sort the data then select the batch sizes from that. (Data Loader). Any Suggestions Please
U should write a custom collate_fn
and sort the data inside it, below is an example:
def collate_fn(batch):
inp, labels = zip(*batch)
inp = SORTED_VERSION
return pad_sequence(inp), tc.stack(labels)
(I assume each index from ur dataset returns a tuple with 2 tensors)
Thank you, but I think bucketing will work here otherwise I wont be able to shuffle the dataset. Do you have any link/suggestion @G.M