Manage Vertex ML Metadata | Vertex AI | Google Cloud
Review Questions
Q1 - B, D ✅
Q2 - C ✅
Q3 - A, D ✅
Q4 - A (Input features' statistical properties change) ✅
Q5 - D ✅
Q6 - C ✅
Q7 - D ❌ B
- This is the definition of prediction drift
- Prediction drift happens when the distribution or patterns of a model’s predicted outputs change over time in production. This drift usually signals that the model is encountering different conditions than it was originally trained on, causing shifts in how it predicts.
Q8 - A ✅
Q9 - A, D ✅
Q10 - D ❌ B, D
- For prediction drift, you will need a baseline statistical distribution of input features of the production data for reference to compare with the continuous statistical distribution of the inputs in production over time
Q11 - C ❌ A
- L-infinity distance is the greatest distance between 2 vectors
- Jensen-Shannon Divergence is for numerical features
Q12 - B, D ❌, B