Machine Learning Operations (MLOps)
What Is MLOps?
MLOps provides critical capabilities to enable machine learning in production, including:
1. Simplified model deployment. Data scientists use a variety of modeling languages, frameworks, and tools. With MLOps, IT operations teams can quickly deploy models from a variety of languages and frameworks in production environments.
2. Monitoring for machine learning. Tools for monitoring software do not work for machine learning. MLOps provides monitoring designed for machine learning. Key capabilities include data drift detection for important features and model-specific metrics.
3. Production life cycle management. The initial model deployment is the beginning of a long life cycle of updates to keep a machine learning model running. MLOps provides a means to test and update models in production without interrupting service to business applications.
4. Production model governance. Machine learning models used in production applications will need to be tightly controlled to prevent unwanted changes and to comply with regulations. MLOps provides access control, traceability, and audit trails to minimize risk and ensure regulatory compliance.
Why Is MLOps Important?
In order to finally realize the value of machine learning, machine learning models must run in production and support efforts to make better decisions or improve efficiency in business applications. Machine learning operations provides the technology and practices to deploy, monitor, manage, and govern machine learning in production. MLOps is required to scale the number of machine learning-driven applications in an organization. MLOps also builds trust through automated processes, testing, and validation that creates a repeatable process for managing machine learning in dynamic environments. It also frees up data scientists to do their own work by empowering the MLOps engineers to take ownership of and responsibility for managing machine learning in production.
MLOps + DataRobot
DataRobot MLOps is a product that is available as part of the DataRobot AI Cloud platform. With DataRobot MLOps, Data Scientists or IT operators can import models built using modern languages like Python, R, Scala, Java, and Go, as well as from most ML platforms available today. The system includes pre-built environments for frameworks, including Keras, Java, PyTorch, and XGBoost, to simplify deployment.
Models are then tested and deployed on Kubernetes and most leading ML execution environments in the market for maximum scalability and performance and are also available via a production-grade REST endpoint. DataRobot MLOps monitoring provides service health, data drift, and accuracy monitoring, reports, and alerts for overall machine learning performance.
Model life cycle management allows models to be updated without interrupting service to downstream applications. DataRobot MLOps also provides robust production model governance with role-based access control, built-in model approval workflows, and full version control and rollback.