How to Shuffle Data in PyTorch DataLoader

To shuffle data in PyTorch's DataLoader, set the parameter shuffle=True when creating the DataLoader. This ensures the data is randomly ordered each epoch, which helps improve model training by reducing bias from data order.

Syntax

The DataLoader class in PyTorch has a shuffle parameter that controls whether the data is shuffled each time it is loaded. Setting shuffle=True randomizes the order of data samples.

• dataset: The dataset to load data from.
• batch_size: Number of samples per batch.
• shuffle: Set to True to shuffle data every epoch, False to keep order.

python

torch.utils.data.DataLoader(dataset, batch_size=1, shuffle=False, ...)

Example

This example shows how to create a DataLoader with shuffled data from a simple dataset of numbers. Each epoch prints batches in a different random order.

python

import torch
from torch.utils.data import DataLoader, TensorDataset

# Create a simple dataset of numbers 0 to 9
data = torch.arange(10)
dataset = TensorDataset(data)

# DataLoader with shuffle=True
loader = DataLoader(dataset, batch_size=3, shuffle=True)

print('Shuffled batches:')
for epoch in range(2):
    print(f'Epoch {epoch + 1}:')
    for batch in loader:
        print(batch[0].tolist())

Output

Shuffled batches:
Epoch 1:
[7, 0, 3]
[1, 9, 6]
[8, 5, 4]
[2]
Epoch 2:
[4, 0, 3]
[9, 7, 6]
[1, 5, 8]
[2]

Common Pitfalls

One common mistake is forgetting to set shuffle=True, which causes the data to be loaded in the same order every epoch. This can lead to poor model generalization because the model sees data in a fixed sequence.

Another pitfall is using shuffle=True with batch_size=1 in very small datasets, which might not show much difference.

python

from torch.utils.data import DataLoader, TensorDataset
import torch

data = torch.arange(5)
dataset = TensorDataset(data)

# Wrong: shuffle=False (default)
loader_no_shuffle = DataLoader(dataset, batch_size=2)
print('No shuffle batches:')
for batch in loader_no_shuffle:
    print(batch[0].tolist())

# Right: shuffle=True
loader_shuffle = DataLoader(dataset, batch_size=2, shuffle=True)
print('Shuffle batches:')
for batch in loader_shuffle:
    print(batch[0].tolist())

Output

No shuffle batches:
[0, 1]
[2, 3]
[4]
Shuffle batches:
[3, 0]
[4, 1]
[2]

Quick Reference

Remember these tips when shuffling data in PyTorch DataLoader:

• Use shuffle=True to randomize data order each epoch.
• Shuffling helps prevent model overfitting to data order.
• For validation or test sets, keep shuffle=False to maintain consistent evaluation.
• Shuffling works only if the dataset supports indexing.

Key Takeaways

Set shuffle=True in DataLoader to randomize data order each epoch.

Shuffling improves model training by reducing bias from fixed data order.

Keep shuffle=False for validation and test DataLoaders to ensure consistent evaluation.

Shuffling requires datasets that support indexing.

Without shuffle=True, data loads in the same order every epoch.