Module 12 Self-Test

A: BERT gives contextual embeddings (same word → different vector based on context); Word2Vec gives one fixed vector per word.

A: Statistical properties don't change over time. Test with ADF (Augmented Dickey-Fuller). Fix with differencing.

A: Model performance degrades as real-world data changes. Data drift (input distribution changes) and concept drift (input-output relationship changes).

A: (a) Churned users were low-value, remaining users spend more. (b) Price increase drove away price-sensitive users.

A: UNION ALL — no deduplication step needed, no sorting/comparing.

A: Halves. SE = σ/√400 = σ/20 vs σ/√100 = σ/10. √n doubled, so SE halved.