Answers

After putting thought into the questions, compare your answers to ours:

1. Delta’s ability to time travel helps with reproducibility. Delta has versioning that allows us a point-in-time lookup to see what data our model was trained on. For long-term versioning, deep clones or snapshots are appropriate.
2. Writing data to an online store provides real-time feature lookup for real-time inference models.
3. Create FeatureLookups for each feature table you wish to include. Then, use create_training_set.
4. A feature table has a unique primary key, which indicates the object or entity the features describe.
5. The possibilities are vast. An example is behavior modeling or customer segmentation to support flagging fraud.