Epoch-based training in PyTorch - Practice Problems & Coding Challenges

Challenge - 5 Problems

❓ Predict Output

intermediate

Output of a simple epoch training loop

What will be the printed output after running this PyTorch training loop for 2 epochs?

Assume the model, optimizer, and loss function are correctly defined and the dataloader has 3 batches with dummy data.

Code:

PyTorch

for epoch in range(2):
    total_loss = 0
    for batch in range(3):
        loss = batch + 1  # dummy loss values: 1, 2, 3
        total_loss += loss
    print(f"Epoch {epoch+1}, Loss: {total_loss}")

A

Epoch 1, Loss: 6
Epoch 2, Loss: 6

B

Epoch 1, Loss: 3
Epoch 2, Loss: 3

C

Epoch 1, Loss: 1
Epoch 2, Loss: 2

D

Epoch 1, Loss: 6
Epoch 2, Loss: 9

💡 Hint

Sum the dummy loss values for each batch inside the epoch.

The dummy loss values per batch are 1, 2, and 3. Their sum is 6. This sum is printed for each epoch, so both epochs print 6.

❓ Model Choice

intermediate

Choosing the correct place to reset loss in epoch training

In an epoch-based training loop, where should you reset the total loss variable to zero to correctly track loss per epoch?

A Before the start of each batch loop

B Before the start of each epoch loop

C After the end of all epochs

D Inside the optimizer step

💡 Hint

Think about when you want to start fresh loss tracking for each epoch.

Resetting total loss before each epoch ensures loss is accumulated fresh for that epoch only.

❓ Hyperparameter

advanced

Effect of increasing number of epochs on model training

What is the most likely effect of increasing the number of epochs during training of a neural network?

A Model may overfit the training data if epochs are too high

B Model will always perform better on unseen data

C Training time decreases as epochs increase

D Loss will increase on training data as epochs increase

💡 Hint

Think about what happens if the model learns too much from training data.

Too many epochs can cause the model to memorize training data, reducing its ability to generalize to new data.

❓ Metrics

advanced

Calculating average loss per epoch

Given a training loop that sums batch losses in total_loss and processes 5 batches per epoch, how do you calculate the average loss per epoch?

A Multiply total_loss by number of batches (5)

B Subtract number of batches (5) from total_loss

C Add number of batches (5) to total_loss

D Divide total_loss by number of batches (5)

💡 Hint

Average means total divided by count.

Average loss per epoch is total loss divided by number of batches processed in that epoch.

🔧 Debug

expert

Identifying the bug in epoch training loop

What error will this PyTorch epoch training loop raise?

Code:
for epoch in range(3): total_loss = 0 for batch in dataloader: optimizer.zero_grad() outputs = model(batch[0]) loss = criterion(outputs, batch[1]) loss.backward() optimizer.step() total_loss += loss print(f"Epoch {epoch+1} Loss: {total_loss}")

A RuntimeError: loss.backward() called multiple times

B NameError: name 'optimizer' is not defined

C TypeError: unsupported operand type(s) for +=: 'int' and 'Tensor'

D No error, code runs correctly

💡 Hint

Check the type of loss and total_loss before adding.

loss is a tensor, total_loss is an int. Adding them directly causes a TypeError.