An ML engineer wants to use Amazon SageMaker Data Wrangler to perform preprocessing on a dataset. The ML engineer wants to use the processed dataset to train a classification model. During…

Question

Options

• APerform ordinal encoding to represent categories of the feature.
• BPerform similarity encoding to represent categories of the feature.
• CPerform one-hot encoding to represent categories of the feature.
• DPerform target encoding to represent categories of the feature.

• A
  4% (2)
• B
  73% (37)
• C
  16% (8)
• D
  8% (4)

Explanation

The key challenge here is that thousands of values differ only by spelling errors - meaning 'sneaker', 'sneeker', and 'sneakar' should be treated as the same category. Similarity encoding represents each category as a vector of character n-gram similarities to all other categories, so misspelled variants of the same word end up with very similar (nearly identical) encodings. This is precisely designed for high-cardinality, noisy categorical text. One-hot encoding (C) would create thousands of orthogonal binary columns, treating each spelling variation as a completely distinct category - defeating the purpose. Ordinal encoding (A) assigns arbitrary integers with no semantic meaning. Target encoding (D) uses target-variable statistics and does not account for string similarity between category values.

No community discussion yet for this question.