Engineering / advanced
Keep online features correct and timely
Connect event history, point-in-time joins, freshness, materialization, and serving contracts to the features models actually consume.
Your route through engineering
Follow the path or begin with a question. Every lesson is open.
Your place, at your pace. Progress is optional and saved only in this browser when available.
Read / explore / reason
Inside this module
Module 01 advanced
Keep online features correct and timely
Connect event history, point-in-time joins, freshness, materialization, and serving contracts to the features models actually consume.
- 01
Feature definitions and entity keys
Specify semantic identities, typed join keys, units, time boundaries, and versioned transformations before storing model inputs.
25 min - 02
Point-in-time correct training joins
Reconstruct eligible feature observations using both event time and actual availability, without leaking later corrections into past decisions.
25 min - 03
Event-time feature aggregation
Define window membership independently of processing order, then specify watermark, trigger, lateness, and correction behavior.
25 min - 04
Online and offline feature parity
Compare the same versioned feature definition at equivalent logical inputs, including time, missingness, preprocessing, and numerical rules.
25 min - 05
Feature materialization and backfill
Populate a new feature generation without mixing revisions, missing recent updates, or overwriting newer state with old backfill results.
25 min - 06
Feature freshness and TTL semantics
Distinguish observation age, availability delay, cache lifetime, source completeness, and feature-specific validity policies.
25 min - 07
Late arrivals and feature corrections
Apply event revisions without duplicate contributions, regressing newer state, or erasing the history needed for as-known retrieval.
25 min - 08
Feature-store hot-key mitigation
Reduce concentrated load while preserving entity semantics, exact aggregate combination, and ordered correction behavior.
25 min - 09
Feature schema compatibility rollouts
Stage producer and consumer versions together, preserving units, defaults, key identity, transformation artifacts, and rollback data.
25 min - 10
Feature lineage and model-consumer audits
Trace versioned source defects to potentially affected models, then use temporal usage evidence to distinguish reachability from observed impact.
25 min