An ML engineer normalized training data by using min-max normalization in AWS Glue DataBrew. The ML engineer must normalize the production inference data in the same way as the training data before…

Question

Options

• AApply statistics from a well-known dataset to normalize the production samples.
• BKeep the min-max normalization statistics from the training set. Use these values to normalize the
• CCalculate a new set of min-max normalization statistics from a batch of production samples. Use
• DCalculate a new set of min-max normalization statistics from each production sample. Use these

• A
  2% (1)
• B
  90% (43)
• C
  6% (3)
• D
  2% (1)

Explanation

Option B is correct because min-max normalization computes (x - min) / (max - min) using fixed reference values - if you recompute those statistics on new data, the scale shifts, and the model receives inputs that no longer match the distribution it was trained on, breaking predictions.

• A is wrong because using statistics from a "well-known dataset" (not your training data) introduces an arbitrary scale that has no relationship to what the model learned.
• C is wrong because recalculating stats from a batch of production samples creates a moving target - the normalization changes with every batch, making inference inconsistent and incomparable to training.
• D is wrong for the same reason as C, but worse: normalizing each sample individually (e.g., a single record with min = max) is mathematically undefined and destroys all relative meaning.

Memory tip: Think of normalization stats as part of the model artifact - they must be frozen at training time and reused at inference time, just like the model weights themselves. "Fit on train, transform on everything."

No community discussion yet for this question.