12 — Pipeline & Automation

Feature Stores#

Data✓ Mathematical
◆ The PatternCentralized feature management with online/offline consistency

A feature store serves two stores from one source of truth: the offline store (historical data for training) and the online store (low-latency data for serving). This solves the training-serving skew problem — the features your model trains on are identical to what it sees in production.

// Interactive — feature store architecture
# Python — Feast feature store
from feast import FeatureStore

store = FeatureStore(repo_path=".")

# Online serving — low latency
features = store.get_online_features(
    features=["user_stats:avg_purchase",
              "user_stats:visit_count"],
    entity_rows=[{"user_id": 123}]
).to_dict()

# Offline training — point-in-time correct
training_df = store.get_historical_features(
    entity_df=entity_df,
    features=["user_stats:avg_purchase"]
).to_df()
Pattern bridge: Feature stores solve training-serving skew just as survivorship bias prevention solves backtesting skew — both ensure the data you test on matches reality.
← Previous
ML Pipelines
Open in the full reader, with the topic sidebar →