Chapter 2: Feature Engineering at Scale

Feature Stores

Feature stores centralize feature computation and serving, ensuring consistency between training and inference.

Feast Example

from feast import FeatureStore, Entity, Feature, ValueType, FeatureView
from feast.infra.offline_stores.file_source import FileSource

# Define entity
customer = Entity(
    name="customer_id",
    value_type=ValueType.STRING,
    description="Customer identifier"
)

# Define feature view
customer_features = FeatureView(
    name="customer_features",
    entities=["customer_id"],
    ttl=timedelta(days=1),
    features=[
        Feature(name="total_purchases", dtype=ValueType.FLOAT),
        Feature(name="days_since_last_login", dtype=ValueType.INT64),
        Feature(name="account_age_days", dtype=ValueType.INT64),
    ],
    online=True,
    source=FileSource(path="data/customer_features.parquet")
)

# Get features for training
store = FeatureStore(repo_path=".")
training_df = store.get_historical_features(
    entity_df=entity_df,
    features=["customer_features:total_purchases", "customer_features:days_since_last_login"]
).to_df()

# Get features for inference (online)
online_features = store.get_online_features(
    features=["customer_features:total_purchases"],
    entity_rows=[{"customer_id": "CUS-12345"}]
).to_dict()

Online vs Offline Features

Type	Purpose	Latency	Storage
Offline	Training	Minutes	Data warehouse
Online	Inference	Milliseconds	Redis, DynamoDB

Key Takeaways

Summary

1. Feature stores ensure training-serving consistency
2. Separate online/offline stores for different latency needs
3. Version features like code and data

---

Next: Chapter 3: Data Quality and Validation