Skip to content

Tecton Feature Platform

The following diagram shows the components of the Tecton Feature Platform.

Feature Store Components

The capabilities of each component of the platform are explained below.

Feature repository

Define and manage features as code

  • Manage features as files in a git repository using a declarative framework

  • Deploy to production with confidence by integrating CI/CD processes in the feature lifecycle

  • Unit test your features before you deploy to production

  • Isolate workflows in separate production and development workspaces

  • Manage dependencies of features across models and version-control features

Feature pipelines

Transform

  • Automate the orchestration and processing of transformation pipelines to materialize feature values

  • Define streaming and real-time transformations in the same way you define batch transformations

  • Run efficient pre-engineered window aggregation pipelines on batch, streaming and real-time features with a single line of code

Feature store

Store

  • Store consistent feature values across offline and online environments to prevent training- serving skew

  • Use an offline store to optimize for large scale, low-cost retrieval for training

  • Use an online store for low-latency retrieval for online serving

Serve

  • Serve features at ultra low-latency with a REST API and scale to over 100,000 QPS

  • Generate accurate training data through a simple Python SDK

  • Automatically backfill features to generate complete training sets

  • Eliminate data leakage through correct time-travel across training and serving environments

Monitor

  • Observe feature availability and freshness

  • Monitor the health of machine learning pipelines and automatically resolve issues that could produce stale feature data

  • Control costs by tracking the computation and storage cost of individual features

Global platform capabilities

Govern

  • Standardize data workflows across the organization

  • Control access to feature data and ensure compliance in your ML applications

Collaborate

  • Discover features through an intuitive Web UI and re-use features to build production models with a single line of code