DataLoader basics in PyTorch - Practice Problems & Coding Challenges

Challenge - 5 Problems

❓ Predict Output

intermediate

What is the output length of batches from this DataLoader?

Given the following PyTorch DataLoader setup, how many batches will it produce?

PyTorch

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

data = torch.arange(10)
dataset = TensorDataset(data)
dataloader = DataLoader(dataset, batch_size=3, shuffle=False)

batches = list(dataloader)
print(len(batches))

A 3

B 4

C 5

D 10

💡 Hint

Think about how many full batches of size 3 fit into 10 items, and what happens to the leftover items.

With 10 items and batch size 3, you get 3 full batches (3*3=9) plus 1 batch with the leftover 1 item, totaling 4 batches.

❓ Model Choice

intermediate

Which DataLoader option enables random shuffling of data each epoch?

You want your model to see data in a different order every time you train. Which DataLoader parameter should you set?

A drop_last=True

B batch_size=1

C num_workers=4

D shuffle=True

💡 Hint

Think about which option changes the order of data samples.

Setting shuffle=True makes DataLoader shuffle the dataset at the start of each epoch.

❓ Hyperparameter

advanced

What effect does setting drop_last=True have in DataLoader?

Consider a dataset with 10 samples and batch_size=4. What happens if drop_last=True?

A The last batch with fewer than 4 samples is dropped, so only 2 batches are returned.

B The last batch is padded with zeros to reach batch size 4.

C All batches are returned, including the last smaller batch.

D DataLoader raises an error if the last batch is smaller than batch size.

💡 Hint

Think about what happens to incomplete batches when drop_last is True.

drop_last=True causes DataLoader to drop the last batch if it has fewer samples than batch_size.

🔧 Debug

advanced

Why does this DataLoader code raise a RuntimeError?

Examine the code below. Why does it raise a RuntimeError about workers?

PyTorch

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

data = torch.arange(5)
dataset = TensorDataset(data)
dataloader = DataLoader(dataset, batch_size=2, num_workers=2)

for batch in dataloader:
    print(batch)

A TensorDataset cannot be used with num_workers > 0.

B batch_size must be 1 when using num_workers > 0.

C num_workers > 0 requires the code to be inside a 'if __name__ == "__main__"' block on Windows.

D DataLoader does not support iteration with num_workers > 0.

💡 Hint

Consider platform-specific multiprocessing rules in Python.

On Windows, using num_workers > 0 requires the DataLoader code to be inside a 'if __name__ == "__main__"' guard to avoid RuntimeError.

🧠 Conceptual

expert

What is the main advantage of using DataLoader with multiple workers?

Why would you set num_workers > 0 in a DataLoader? Choose the best explanation.

A It allows loading data in parallel, reducing waiting time during training.

B It increases the batch size automatically for faster training.

C It guarantees data is shuffled more thoroughly each epoch.

D It compresses data to save memory during loading.

💡 Hint

Think about how multiple workers affect data loading speed.

Multiple workers load data in parallel, which can speed up training by reducing data loading bottlenecks.