[FIXED] Pytorch DataLoader shuffle=False?

Issue

I used Pytorch DataLoader to create My "batch-data" loder,but I got some problem.

As the definition of the pytorch DataLoader Shuffer.

shuffle (bool, optional) – set to True to have the data reshuffled at every epoch (default: False)

the data will be reshuffled after every epoch. But,though I set shuffle to False,I will probably also get the completely different batch every iteration in the same epoch which I expect .

testData = torchvision.datasets.FashionMNIST(
    root="data",
    train=False,
    download=True,
    transform=ToTensor()
)

CurrentFoldTestDataLoader = data.DataLoader(testData, batch_size=32, shuffle=False)
for i in range(1000):
    test_features, test_labels = next(iter(CurrentFoldTestDataLoader))
    print(i,test_labels)

Here I got the same batch in every iteration.

0 tensor([9, 2, 1, 1, 6, 1, 4, 6, 5, 7, 4, 5, 7, 3, 4, 1, 2, 4, 8, 0, 2, 5, 7, 9,
        1, 4, 6, 0, 9, 3, 8, 8])
1 tensor([9, 2, 1, 1, 6, 1, 4, 6, 5, 7, 4, 5, 7, 3, 4, 1, 2, 4, 8, 0, 2, 5, 7, 9,
        1, 4, 6, 0, 9, 3, 8, 8])
2 tensor([9, 2, 1, 1, 6, 1, 4, 6, 5, 7, 4, 5, 7, 3, 4, 1, 2, 4, 8, 0, 2, 5, 7, 9,
        1, 4, 6, 0, 9, 3, 8, 8])
3 tensor([9, 2, 1, 1, 6, 1, 4, 6, 5, 7, 4, 5, 7, 3, 4, 1, 2, 4, 8, 0, 2, 5, 7, 9,
        1, 4, 6, 0, 9, 3, 8, 8])
4 tensor([9, 2, 1, 1, 6, 1, 4, 6, 5, 7, 4, 5, 7, 3, 4, 1, 2, 4, 8, 0, 2, 5, 7, 9,
        1, 4, 6, 0, 9, 3, 8, 8])
5 tensor([9, 2, 1, 1, 6, 1, 4, 6, 5, 7, 4, 5, 7, 3, 4, 1, 2, 4, 8, 0, 2, 5, 7, 9,
        1, 4, 6, 0, 9, 3, 8, 8])
6 tensor([9, 2, 1, 1, 6, 1, 4, 6, 5, 7, 4, 5, 7, 3, 4, 1, 2, 4, 8, 0, 2, 5, 7, 9,
        1, 4, 6, 0, 9, 3, 8, 8])
7 tensor([9, 2, 1, 1, 6, 1, 4, 6, 5, 7, 4, 5, 7, 3, 4, 1, 2, 4, 8, 0, 2, 5, 7, 9,
        1, 4, 6, 0, 9, 3, 8, 8])
8 tensor([9, 2, 1, 1, 6, 1, 4, 6, 5, 7, 4, 5, 7, 3, 4, 1, 2, 4, 8, 0, 2, 5, 7, 9,
        1, 4, 6, 0, 9, 3, 8, 8])
9 tensor([9, 2, 1, 1, 6, 1, 4, 6, 5, 7, 4, 5, 7, 3, 4, 1, 2, 4, 8, 0, 2, 5, 7, 9,
        1, 4, 6, 0, 9, 3, 8, 8])
10 tensor([9, 2, 1, 1, 6, 1, 4, 6, 5, 7, 4, 5, 7, 3, 4, 1, 2, 4, 8, 0, 2, 5, 7, 9,
        1, 4, 6, 0, 9, 3, 8, 8])

Why is this? Is my understanding of the definition of shuffle inaccurate?

Solution

The problem with your code is that you are re-instantiating the same iterator for each step in the for cycle. With shuffle=False the iterator generates the same first batch of images. Try to instantiate the loader outside the cycle instead:

loader = data.DataLoader(testData, batch_size=32, shuffle=False)

for i, data in enumerate(loader):
    test_features, test_labels = data
    print(i, test_labels)

Answered By - aretor