Train/val/test split in PyTorch - Practice Problems & Coding Challenges

Challenge - 5 Problems

❓ Predict Output

intermediate

Output of PyTorch dataset split code

What will be the output of the following code snippet that splits a dataset into train, validation, and test sets using PyTorch's random_split?

PyTorch

from torch.utils.data import random_split
from torch.utils.data import Dataset

class DummyDataset(Dataset):
    def __init__(self, length):
        self.data = list(range(length))
    def __len__(self):
        return len(self.data)
    def __getitem__(self, idx):
        return self.data[idx]

full_dataset = DummyDataset(100)
train_size = 70
val_size = 20
test_size = 10
train_set, val_set, test_set = random_split(full_dataset, [train_size, val_size, test_size])

print(len(train_set), len(val_set), len(test_set))

A 33 33 34

B 100 0 0

C 70 20 10

D 50 25 25

💡 Hint

Check how the sizes are passed to random_split and what lengths are printed.

The random_split function divides the dataset into subsets with the exact sizes provided. Here, 70, 20, and 10 sum to 100, so the lengths printed are 70, 20, and 10 respectively.

🧠 Conceptual

intermediate

Purpose of validation set in train/val/test split

What is the main purpose of the validation set in a train/val/test split during machine learning model training?

A To tune hyperparameters and prevent overfitting

B To train the model with more data

C To evaluate the model's performance on unseen data after training

D To test the model's final accuracy before deployment

💡 Hint

Think about which set helps decide model settings without biasing final evaluation.

The validation set is used during training to tune hyperparameters and monitor if the model is overfitting. It is separate from the test set, which is used only for final evaluation.

❓ Hyperparameter

advanced

Choosing split ratios for train/val/test

Which of the following split ratios is most appropriate for a dataset with 10,000 samples to ensure reliable training, validation, and testing?

A 60% train, 20% validation, 20% test

B 50% train, 25% validation, 25% test

C 90% train, 5% validation, 5% test

D 80% train, 10% validation, 10% test

💡 Hint

Consider enough data for training and meaningful validation/testing.

80/10/10 is a common split that balances enough training data with sufficient validation and test samples for reliable evaluation.

🔧 Debug

advanced

Identify error in PyTorch dataset splitting code

What error will the following code raise when trying to split a dataset into train, validation, and test sets?

PyTorch

from torch.utils.data import random_split
full_dataset = list(range(50))
train_set, val_set, test_set = random_split(full_dataset, [30, 15, 10])

A TypeError: 'list' object has no attribute '__len__'

B ValueError: Sum of input lengths does not equal the length of the input dataset

C TypeError: random_split expects a Dataset object, not a list

D No error, splits successfully

💡 Hint

Check the sum of split sizes compared to dataset length.

The sum of split sizes is 30+15+10=55, which is greater than the dataset length 50, causing a ValueError.

❓ Model Choice

expert

Best approach to split highly imbalanced dataset

You have a highly imbalanced classification dataset with 1% positive and 99% negative samples. Which approach is best to split the dataset into train, validation, and test sets to maintain class distribution?

A Use stratified splitting to keep class proportions in each subset

B Manually shuffle and split the dataset randomly

C Use random_split from PyTorch directly without stratification

D Split only into train and test, skip validation

💡 Hint

Think about preserving class balance in each subset for fair evaluation.

Stratified splitting ensures each subset has the same class distribution as the full dataset, which is crucial for imbalanced data.