feature-store

name	feature-store
description	Design and manage feature stores for ML systems. Use when building reusable feature pipelines, managing feature versioning, or serving features for inference.

Feature Store

Activate this skill when working with feature engineering at scale.

When to Use

• Building reusable feature computation pipelines
• Managing feature versioning and lineage
• Serving features for online inference
• Sharing features across ML models
• Ensuring training-serving consistency

Tools

• Feast: Open-source feature store
• Hopsworks: Full-featured platform
• Custom: pandas + SQL + caching

Patterns

from feast import FeatureStore, Entity, FeatureView

store = FeatureStore(repo_path="feature_repo/")

# Define features
user_features = FeatureView(
    name="user_features",
    entities=[user_entity],
    schema=[
        Field(name="total_purchases", dtype=Int64),
        Field(name="avg_order_value", dtype=Float64),
    ],
    source=user_source,
)

# Retrieve for training
training_df = store.get_historical_features(
    entity_df=entity_df,
    features=["user_features:total_purchases"],
).to_df()

Rules

• Compute features once, use everywhere
• Version features alongside model versions
• Monitor feature distributions for drift
• Document feature semantics and business logic
• Test feature pipelines with known inputs