Feature Engineering Quiz

Answers and explanations appear as you take the quiz.

1. What work does feature engineering involve before model training?
   • Choosing the fastest GPU for training
   • Writing the model's loss function
   • Transforming raw data into informative inputs a model can learn from
   • Deploying a model to production
2. Why should a proposed engineered feature be evaluated with the chosen model and task?
   • Every extra feature guarantees higher accuracy
   • Its usefulness depends on what it represents and how the model uses it
   • Input representation never affects predictions
   • Changing features always matters more than every algorithm choice
3. A feature uses an outcome recorded only after the prediction is needed. Which feature-engineering problem does this illustrate?
   • A necessary improvement in causal identification
   • A harmless change to a file extension
   • A requirement to use more graphics hardware
   • Prediction-time information leakage
4. When fitting a standardization step for a train-and-held-out evaluation, which data should supply its mean and scale?
   • The training data, with those fitted statistics reused on held-out inputs
   • The test data only
   • All data after examining final test outcomes
   • A different refit on every evaluation example
5. Which of these is a typical feature engineering technique for categorical data?
   • Increasing the learning rate
   • One-hot or target encoding
   • Adding more GPUs
   • Shuffling the file names
6. With OneHotEncoder configured as handle_unknown ignore and no dropped category, how is an unseen category represented for its feature?
   • An extra indicator column created automatically for that new category
   • An immediate error regardless of the stated ignore setting
   • All zeros in the indicator columns for that feature
   • The outcome target value copied into the indicator columns
7. Why can extreme numeric observations affect a StandardScaler transformation?
   • Its estimated mean and variance are sensitive to outliers
   • It uses medians and interquartile ranges instead of means and variances
   • It ranks observations without using their numeric magnitudes
   • It clips every value to fitted quantiles before estimating scale
8. In a stated degree-two interaction feature x1 times x2, what is its value for x1 three and x2 four?
   • Seven
   • One
   • Twenty-five
   • Twelve