Is it possible to fetch multiple features from feast feature-store in a single request?

Question

In use-cases like search one might need to apply a model to multiple entries at runtime. Having a single query is much more resilient than one per product. Does feast support it?

Answer 1

In this scenario, the "feature service" component of the Feast can be utilized to pull data from one or more feature views. This makes it possible to fetch multiple features in a single call.

For example, let's say we have multiple feature views -

1. Feature view for demographics data

demographics_fv = FeatureView(
    name="demographics_fv",
    entities=[cust_id],
    ttl=timedelta(days=365),
    schema=[
        Field(name="gender", dtype=String),
        Field(name="age", dtype=Int32),
        Field(name="location", dtype=String),
    ],
    online=True,
    source=cust_data,
    tags={}
)

2. Feature view for medical data

medical_fv = FeatureView(
    name="medical_fv",
    entities=[cust_id],
    ttl=timedelta(days=365),
    schema=[
        Field(name="weight", dtype=Int32),
        Field(name="height", dtype=Int32),
        Field(name="blood_group", dtype=String),
    ],
    online=True,
    source=cust_data,
    tags={}
)

A feature service definition can be created that will consist references to multiple feature views -

cust_data_fs = FeatureService(
    name="cust_data_fs",
    features=[demographics_fv, medical_fv[["weight", "height"]]]
)

Now a call can be made to this feature service to retrieve required data that may be coming from one or more feature views -

feature_service = feature_store.get_feature_service("cust_data_fs")
feature_store.get_historical_features(features=feature_service, entity_df=entity_df)

You can find more details on this link: https://docs.feast.dev/getting-started/concepts/feature-retrieval#feature-services