PRODUCTION SCENARIO
A churn model scores every customer on the first day of the month. One training dataset spans 24 months and is built in SageMaker Feature Store by joining each customer to their most recent feature record, including updates written later in the same month.
Which create_dataset option removes the future information from the training features?
Answering here is anonymous. Nothing is saved unless you sign in.
Show answer and explanation
Answer: point_in_time_accurate_join, so each row reads only records older than its event time
A point-in-time accurate join bounds each row by its own event time, so a customer scored in March never sees an update written in April. as_of applies one global cutoff to the whole dataset, which cannot vary per row across 24 months. An event time range selects which months appear rather than which updates each row sees, and keeping one record per customer still keeps the most recent one.