Feature Engineering Checkpoint

Reuse fitted transforms, handle constant columns, and preserve categorical column meaning. Select one answer per multiple-choice question.

Get every question correct to pass this checkpoint. You can review feedback, revise your answers, and submit again.

1. Training minimum is 4 and maximum is 8. Using the saved min-max transform without clipping, what is the transformed held-out value 10? Enter a decimal.

2. A column was always 7 in training; its next value is 9. What do the exercise min-max and standardization transforms return, respectively?

Select one answer.

0 and 0 2 and 2 0 and 2 1 and 1

3. Training values are [2,6]. Which mean and effective scale does standardize return for this column?

Select one answer.

Mean 4, scale 2 Mean 4, scale sqrt(8) Mean 8, scale 4 Mean 4, scale 1

4. A scaler is fitted to all rows before a train/test split, using no labels. What is the problem?

Select one answer.

No problem, because only target labels can leak No problem, because scaling cannot affect model fitting The test labels become one-hot encoded Test feature distributions influence preprocessing fitted for training

5. Saved categories are [bird,cat,dog]. Which vector encodes cat?

Select one answer.

[1,0,0] [0,1,0] [0,0,1] [1,1,0]

6. Training categories are [bird,cat,dog], but a prediction batch contains only [dog,cat]. Which transformation preserves the model input?

Select one answer.

Refit one_hot_encode on the prediction batch to remove unused columns Sort the prediction rows before encoding them Call encode_value for each new value using the saved three-category list Replace the categories with [dog,cat] to match arrival order

7. With saved categories [bird,cat,dog], the value fox is unseen. Under the exercise policy, what happens?

Select one answer.

It becomes [0,0,0], and decoding that row returns None A fourth column is appended automatically It becomes [1,0,0], the first known category Decoding [0,0,0] recovers fox

8. After standardizing training data, a held-out batch has mean 2 in one transformed column. What must you do?

Select one answer.

Refit the scaler on that batch until its mean is zero Subtract that batch mean before every prediction Reject the batch because standardization guarantees zero mean everywhere Reuse the fitted scaler; held-out data need not have zero mean

Auto-advance on pass

Join Discord