Feature store

As explained in Machine Learning System Design

An infrastructure system for storing, managing, and serving machine-learning features.

Suppose two teams spend weeks calculating the same feature. One implementation is tested but hard to discover; the other is easier to reuse but produces different values. A feature store replaces that fragmented approach with a central hub for calculating, storing, aggregating, testing, documenting, and monitoring features used by ML pipelines. section 11.4 section 11.4

The payoff is shared work: teams can discover existing features, inspect their data sources and calculations, and reuse them across projects. Standardized calculations also reduce discrepancies between models or pipeline stages. Building and managing the store is itself work, so its value depends on how much duplication and inconsistency it removes. section 11.4.1 section 11.4.1

Before designing one, ask:

• How many features must be written and read, and how quickly must reads return? Computing a lightweight feature at request time may be as fast as fetching a stored value. section 11.4.2
• When should features be refreshed? A long job might update them daily, while lightweight features might be streamed near real time. Refreshing recent days can also replace values affected by temporary upstream problems. section 11.4.2
• What can be calculated ahead of time? Precomputation avoids repeating work when gathering a dataset; a daily update could follow the database’s processing of the previous day’s orders. section 11.4.2

Figure 11.13 Online features and batch features are written in a feature store in different ways but are consumed in the same manner.

Online and batch features can take different paths into the store while being consumed in the same manner. This is useful when some predictions need recent events alongside precomputed values. section 11.4.2

Consider a feature store especially when an ML system draws on multiple sources of tabular data. A pure deep-learning problem is a poor fit. Start with the features teams actually need to share, then weigh the operational effort against the time and consistency gained. section 11.6 section 11.4.1

How Machine Learning System Design covers it

How deep it goes

Works through benefits and design questions around read latency, update schedules, and precomputation, without specifying a full implementation.

Examples

Uses duplicated feature implementations, daily order processing, and a diagram contrasting online and batch feature writes.

What it assumes

Assumes familiarity with features and ML pipelines; defines the store’s role rather than teaching feature engineering from zero.

What you can do after

Evaluate/understand it by weighing reuse against management effort and checking read latency, write load, and refresh schedules.

How it connects

Links feature engineering to ML pipelines, batch processing, streaming, and data warehouses.

liveBookmanning.com

Read Machine Learning System Design on liveBook Get the book at manning.com

Also explained in

• 

  liveBookmanning.com
  Machine Learning Platform EngineeringA feature store is a system that centrally stores, manages, and serves curated machine-learning features for training and inference.
• 

  liveBookmanning.com
  LLMs in ProductionA feature store is a system or abstraction for managing, documenting, versioning, governing, and serving machine-learning features for training and production.
• 

  liveBookmanning.com
  Delivering Machine Learning Projects: From design to deploymentA feature store is a system for storing, managing, and serving machine learning features for model development and inference.
• 

  liveBookmanning.com
  Financial AI in PracticeA centralized repository for storing, managing, and serving validated features for machine-learning workflows.

Compare how 5 books explain feature store →

Related concepts

• inputuses · An input is a value, information, or signal supplied to a system, computation, or interface for processing or to influence its behavior.
• training-serving skewaddresses · Training-serving skew is a mismatch between the data, feature representations, or computations used during model training and those used to generate predictions.
• Apache Airflowcontrasts with · Apache Airflow is an open-source platform for authoring, scheduling, and monitoring programmatically defined workflows, commonly represented as directed acyclic graphs.
• machine learninguses · Machine learning is a field of computational methods in which systems learn patterns or relationships from data to make predictions, decisions, or other outputs.
• independent variableuses · An independent variable is an input or explanatory attribute whose values are used to model, explain, or predict variation in an outcome.
• source of truthaddresses · A source of truth is the authoritative source or representation from which the accepted value or state of information is determined.

This explainer was generated by Manning's knowledge graph from the text of Machine Learning System Design. The section links point to the parts of the book it draws on.