Batch size and shuffling in PyTorch - Cheat Sheet & Quick Revision

Recall & Review

beginner

What is batch size in machine learning training?

Batch size is the number of samples processed together before the model updates its weights. It controls how many examples the model sees at once.

beginner

Why do we shuffle data during training?

Shuffling mixes the order of data samples to prevent the model from learning patterns based on the order. It helps the model generalize better.

intermediate

How does a large batch size affect training speed and memory?

A large batch size speeds up training by processing more samples at once but uses more memory. It may also affect how well the model learns.

intermediate

What is the effect of not shuffling data in each training epoch?

Not shuffling can cause the model to see data in the same order every time, which might lead to overfitting or poor generalization.

beginner

In PyTorch, how do you enable shuffling when creating a DataLoader?

You set the argument shuffle=True in the DataLoader constructor to shuffle data at every epoch.

What does batch size control during training?

A Number of layers in the model

B Number of samples processed before updating model weights

C Learning rate of the optimizer

D Number of epochs

Batch size controls how many samples the model processes before updating its weights.

Why is shuffling data important during training?

A To prevent the model from learning order-based patterns

B To reduce memory usage

C To speed up training

D To increase batch size

Shuffling prevents the model from learning patterns based on the order of data.

What happens if batch size is too large?

A Training becomes faster but uses more memory

B Training becomes slower and uses less memory

C Model accuracy always improves

D Model stops training

Large batch sizes speed up training but require more memory.

In PyTorch DataLoader, how do you shuffle data?

A shuffle=False

B num_workers=0

C shuffle=True

D batch_size=1

Setting shuffle=True enables shuffling of data each epoch.

What is a possible risk of not shuffling data during training?

A DataLoader crashes

B Model trains faster

C Batch size increases automatically

D Model may overfit or generalize poorly

Not shuffling can cause the model to overfit or generalize poorly.

Explain what batch size is and how it affects training speed and memory.

Think about how many samples the model sees at once and what happens if you change that number.

Describe why shuffling data is important during training and how to enable it in PyTorch.

Consider what happens if data order stays the same every epoch.