Practice set: features, labels, error, baselines and test data

P1. In a table that predicts house price from area, the label is:

• Price
• Area
• The row number
• Nothing

P2. Values 2, 4, 6, 8 have mean:

• 5
• 4
• 6
• 20

P3. The baseline MAE for 2, 4, 6, 8 (predicting 5) is:

• 2
• 5
• 1
• 8

P4. Predicting 9 when the truth is 12 gives absolute error:

• 3
• −3
• 21
• 108

P5. A model with 100% training accuracy and 40% test accuracy is probably:

• Over-fitted
• Perfect
• Under-trained and safe
• A baseline

P6. The test set should be used:

• Once, at the end
• During every fit
• Instead of training data
• Never

P7. Using the answer as a feature is called:

• Leakage
• Baseline
• Supervision
• Validation

P8. A baseline for a yes/no label usually predicts:

• The most common class
• Always no
• Random answers
• The mean