Train/Validation/Test Split

    Machine Learning · Beginner

    Definition

    Data is split into training (learn), validation (tune) and test (final evaluation) sets.

    Prevents fooling yourself with scores on data the model has seen.

    Diagram

    Train 70%Val 15%Test 15%

    Flowchart

    Shuffle dataSplit 70/15/15Train on trainTune on validationReport on test once

    Example: Simple split

    data = list(range(10))
n = int(len(data) * 0.8)
print(len(data[:n]), len(data[n:]))

    Output: 8 2

    Key points

    • Never tune on the test set
    • Stratify for imbalanced classes

    Interview questions

    What is data leakage?

    When information from the test set leaks into training, inflating scores.

    More Machine Learning topics

    What is Machine LearningLinear RegressionLogistic RegressionDecision Trees & Random Forest