Answers for "pytorch bucket dataloader"

0

pytorch bucket dataloader

from torch.nn.utils.rnn import pack_sequence
from torch.utils.data import DataLoader

def my_collate(batch):
    # batch contains a list of tuples of structure (sequence, target)
    data = [item[0] for item in batch]
    data = pack_sequence(data, enforce_sorted=False)
    targets = [item[1] for item in batch]
    return [data, targets]

# ...
# later in you code, when you define you DataLoader - use the custom collate function
loader = DataLoader(dataset,
                      batch_size,
                      shuffle,
                      collate_fn=my_collate, # use custom collate function here
                      pin_memory=True)
Posted by: Guest on August-27-2021

Python Answers by Framework

Browse Popular Code Answers by Language