Feast
Modeling

Feast

Released: 
Apr 2021
  •  
Documentation
  •  
License:  
Apache-2.0 License
81
Github open issues
2382
Github stars
21 Oct
Github last commit
9
Stackoverflow questions

What is Feast?

Feast is an open-source feature store helping manage and serve machine learning features to models in production. It serves feature data to models using two options:

  1. A low-latency online store for real-time prediction
  2. An offline store for batch scoring or model training

Feast helps define, discover, serve, validate, and manage features to machine learning models during training and online inference.

It provides a single data access layer that abstracts feature storage from feature retrieval and decouples models from data infrastructure. It also helps minimize training-serving skew by ensuring consistency in data fed for training and online inference.


How Does Feast Help?

Feast brings consistency and standardization to data engineering workflows and lays a solid foundation for organizational ML platforms.

It eliminates the need for management and deployment of dedicated infrastructure by reusing current infrastructure and allotting new resources when needed. It enables easy consumption of existing feature views by teams and helps avoid starting from scratch. 

With a centralized registry, Feast aids data science professionals to publish features and helps engineering teams ship features into production with minimal oversight and organizational friction. It also addresses the inconsistencies generated due to data leakage issues with online features retrieval while exporting datasets for model training ensuring point in time correctness. 


Key Features of Feast

Rich data ingestion 

Feast is designed to ingest data from multiple data sources such as object stores, streams, databases, or notebooks. 

Training-serving consistency

Feast offers a consistent view of feature data using a unified serving API and ingestion layer. It eliminates training-serving skew by decoupling models from data infrastructure with a single data access layer.

Standardized integration 

Feast works as a single source of truth for all features for training and data specifications. Feast enables communicating features, testing feature data, determining feature relevance, and tracking metadata like event timestamp, metrics, and documentation.  

Feature sharing

Feast offers a metadata API to enable teams to share, track and reuse historical features across several projects. It executes on top of cloud-managed services and spins up resources readily.

Statistics with validation 

Powered by TFDV compatibility, Feast allows validating statistics generated on the basis of features within the systems. 

Companies using

Feast

salesforce
shopify
gojek
ibm
No items found.

Liked the content? You'll love our emails!

The best MLOps and AI Observability content handpicked and delivered to your email twice a month

Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.

Censius automates model monitoring

so that you can 

boost healthcare

improve models

scale businesses

detect frauds

boost healthcare

improve models

scale businesses

detect frauds

boost healthcare

Start Monitoring