Skip to main content

Feature Definitions

To build production-ready features, Raptor must understand the logic of the Feature and create the engineering setup required for the production version and introduce a new way of describing features - Feature Definitions.

Feature Definitions are resources that bundle the Feature's logic and the engineering aspects; they're usually written in Python using the LabSDK, and exported as Kubernetes manifests and composed of two parts:

  1. Declarative definition-
    Instruct Raptor on implementing the advanced mechanisms in production while implementing them differently for training purposes.
  2. Imperative definition-
    The Python code allows you the freedom and creativity to do fancy stuff.

Feature Definition

Raptor uses these definitions to create your features in research and production environments, provision the necessary data connections, compute and storage resources, and offer collaboration, management and governance layers around features.

Features are usually written by the data-scientists while developing the model using the LabSDK, then they can be exported to the Feature Definitions as YAML files - and are both human and machine-readable.