Early stopping implementation in PyTorch - Practice Problems & Coding Challenges

Challenge - 5 Problems

❓ Predict Output

intermediate

Output of Early Stopping Check Function

Given the following early stopping check function, what will be the output after calling it with the provided inputs?

PyTorch

def early_stopping_check(val_loss, best_loss, patience_counter, patience):
    if val_loss < best_loss:
        best_loss = val_loss
        patience_counter = 0
        return True, best_loss, patience_counter
    else:
        patience_counter += 1
        if patience_counter >= patience:
            return False, best_loss, patience_counter
        else:
            return True, best_loss, patience_counter

best_loss = 0.5
patience_counter = 2
patience = 3
val_loss = 0.6

result = early_stopping_check(val_loss, best_loss, patience_counter, patience)
print(result)

A (False, 0.6, 3)

B (True, 0.5, 3)

C (False, 0.5, 3)

D (True, 0.6, 0)

💡 Hint

Think about what happens when validation loss does not improve and patience counter reaches the limit.

The validation loss (0.6) is not less than the best loss (0.5), so patience_counter increments from 2 to 3. Since patience is 3, the function returns False to stop training.

❓ Model Choice

intermediate

Choosing Early Stopping Patience Value

You are training a neural network and want to use early stopping. Which patience value is most suitable to avoid stopping too early but still prevent overfitting?

A Patience = 5 (stop after 5 bad epochs)

B Patience = 50 (stop after 50 bad epochs)

C Patience = 1 (stop after 1 bad epoch)

D Patience = 0 (stop immediately on first bad epoch)

💡 Hint

Consider a balance between giving the model time to improve and avoiding wasting time on no improvement.

A patience of 5 allows the model some room to improve after a bad epoch but stops training before wasting too much time.

🔧 Debug

advanced

Debugging Early Stopping Implementation

The following early stopping code does not stop training even when validation loss stops improving. What is the bug?

PyTorch

class EarlyStopping:
    def __init__(self, patience=3):
        self.patience = patience
        self.counter = 0
        self.best_loss = float('inf')
        self.early_stop = False

    def __call__(self, val_loss):
        if val_loss >= self.best_loss:
            self.counter += 1
            if self.counter >= self.patience:
                self.early_stop = True
        else:
            self.best_loss = val_loss
            self.counter = 0

# Usage example
es = EarlyStopping(patience=2)
losses = [0.5, 0.4, 0.45, 0.46, 0.47]
for loss in losses:
    es(loss)
print(es.early_stop)

A The early_stop flag is never set to True

B The best_loss is never updated, so early_stop never triggers

C The counter is reset incorrectly inside the else block

D The condition should be val_loss >= self.best_loss instead of val_loss > self.best_loss

💡 Hint

Think about what happens when val_loss equals best_loss.

If val_loss equals best_loss, the counter does not increment, so early stopping may never trigger. Changing > to >= fixes this.

❓ Metrics

advanced

Interpreting Early Stopping Training Logs

During training with early stopping, the validation loss values per epoch are: [0.6, 0.55, 0.54, 0.54, 0.55, 0.56, 0.57]. If patience is set to 2, at which epoch will training stop?

A After epoch 6

B Training will not stop early

C After epoch 7

D After epoch 5

💡 Hint

Count how many consecutive epochs the validation loss does not improve.

Validation loss stops improving after epoch 3 (0.54). It stays the same at epoch 4, then worsens at epochs 5 and 6. After 2 bad epochs (5 and 6), training stops.

🧠 Conceptual

expert

Why Use Early Stopping in Model Training?

Which of the following best explains the main reason to use early stopping during training of machine learning models?

A To reduce training time by stopping as soon as training loss decreases

B To prevent overfitting by stopping training when validation loss stops improving

C To increase model complexity by training longer

D To ensure the model reaches zero training loss

💡 Hint

Think about the difference between training loss and validation loss.

Early stopping monitors validation loss to stop training before the model starts overfitting the training data.