Machine Learning

Built on open lakehouse architecture, Databricks Machine Learning empowers ML teams to prepare and process data, streamlines cross-team collaboration and standardizes the full lifecycle from experimentation to production.

Simplify all aspects of data for ML

Because Databricks ML is built on an open lakehouse foundation with Delta Lake, you can empower your machine learning teams to access, explore and prepare any type of data at any scale. Turn features into production pipelines in a self-service manner without depending on data engineering support.

Automate experiment tracking and governance

Managed MLflow automatically tracks your experiments and logs parameters, metrics, versioning of data and code, as well as model artifacts with each training run. You can quickly see previous runs, compare results and reproduce a past result, as needed. Once you have identified the best version of a model for production, register it to the Model Registry to simplify handoffs along the deployment lifecycle.

Manage the full model lifecycle with the Model Registry

Once trained models are registered, you can collaboratively manage them through their lifecycle with the Model Registry. Models can be versioned and moved through various stages, like experimentation, staging, production and archived. The lifecycle management integrates with approval and governance workflows according to role-based access controls. Comments and email notifications provide a rich collaborative environment for data teams.

Deploy ML models at scale and low latency

From the Model Registry, quickly deploy production models using batch scoring for scale, or Databricks Serving for low-latency online serving as REST API endpoints. Because the Model Registry relies on the MLflow Model format, it benefits from ecosystem integrations for a wide variety of deployments, like deploying Docker containers on Kubernetes or loading a model onto a device.