Feast (Feature Store) is a tool for managing and serving machine learning features. Feast is the bridge between models and data.
Feast aims to:
- Provide a unified means of managing feature data from a single person to large enterprises.
- Provide scalable and performant access to feature data when training and serving models.
- Provide consistent and point-in-time correct access to feature data.
- Enable discovery, documentation, and insights into your features.
TL;DR: Feast decouples feature engineering from feature usage. Features that are added to Feast become available immediately for training and serving. Models can retrieve the same features used in training from a low latency online store in production. This means that new ML projects start with a process of feature selection from a catalog instead of having to do feature engineering from scratch.
# Setting things up
fs = feast.Client('feast.example.com')
customer_features = ['CreditScore', 'Balance', 'Age', 'NumOfProducts', 'IsActive']
# Training your model (typically from a notebook or pipeline)
data = fs.get_batch_features(customer_features, customer_entities)
my_model = ml.fit(data)
# Serving predictions (when serving the model in production)
prediction = my_model.predict(fs.get_online_features(customer_features, customer_entities))
The following commands will start Feast in online-only mode.
git clone https://github.com/feast-dev/feast.git
cd feast/infra/docker-compose
cp .env.sample .env
docker-compose -f docker-compose.yml -f docker-compose.online.yml up -d
This will start a local Feast deployment with online serving. Additionally, a Jupyter Notebook with Feast examples.
Please see the links below to set up Feast for batch/historical serving with BigQuery.
Please refer to the official documentation at https://docs.feast.dev
Feast is a community project and is still under active development. Your feedback and contributions are important to us. Please have a look at our contributing guide for details.