Are you interested in MLOps? You’re not alone. In the past two years, interest in MLOps has exploded:
Today, there are hundreds of vendors who say they support MLOps. But do they really? Like many “hot” technology terms, “MLOps” appears to mean several different things. Let’s clarify.
What is MLOps?
If you ask “who offers an MLOps platform?” you are asking the wrong question. Everyone says they offer MLOps. In today’s market, ask who offers the broadest and most complete MLOps platform.
MLOps is an operational discipline enabling enterprises to deliver machine learning at scale. An MLOps platform drives value through continuous development, testing, deployment, monitoring, and retraining. MLOps platforms accelerate model velocity by automating tasks and streamlining processes. They reduce the total cost of a machine learning program by consolidating tools and managing infrastructure.
Some vendors use the term “MLOps” to refer to the “last mile” of model deployment. This is incomplete. An enterprise-grade MLOps platform supports the complete machine learning lifecycle:
Enterprise-grade MLOps platforms drive value with comprehensive capabilities, including:
- Project management
- Comprehensive asset management
- Model deployment
- Model monitoring
- Model governance
- Data and platform integration
Large organizations use highly diverse tools, data, and computing platforms for machine learning. An enterprise-grade platform works with any computing platform, any data, and any machine learning tool. And, it should meet IT requirements for security and cost management.
Comparing MLOps Market Segments
Vendor offerings for MLOps fall into four broad segments: Big Cloud, Big Data, Low Code/No Code Vendors, and MLOps Specialists.
Big Cloud vendors invest in machine learning tools to drive workloads to their platforms. Their tools integrate well with the other services they offer and appeal to developers.
- Amazon SageMaker
- Google Vertex AI
- Microsoft Azure Machine Learning
Tools from these vendors tend to have low appeal for data scientists. They lack strong collaboration and project management capabilities. And, of course, they are only an option for customers committed to a single computing platform.
Big Data vendors invest in machine learning tools to add value to their data platforms. They generally offer strong data and metadata capture capabilities, and they appeal to data engineers.
- Cloudera Machine Learning
- Databricks Machine Learning
- Oracle Data Science
- Teradata Vantage
Big data tools also have limited appeal for data scientists. Tight integration with a data platform is helpful for customers who are fully committed to that platform. However, in the real world, most data scientists work with data from many different sources and platforms.
Low-Code/No-Code Vendors. These vendors invest in MLOps so that users can deploy the models they develop with their software. They appeal to “citizen data scientists” and users who prefer a no-code interface.
- Alteryx Promote
- Dataiku Data Science Studio
- DataRobot MLOps
These tools lack the flexibility that expert data scientists require. They offer limited MLOps support for models developed with other tools.
MLOps Specialists focus on specific capabilities within the MLOps domain, such as explainability or model monitoring.
- Arize AI (Monitoring and Observability)
- Fiddler AI (Explainability)
- Seldon (MLOps on Kubeflow)
- Tecton (Feature Store)
- TruEra (AI Quality Assurance)
- Weights & Biases (Experiment Management)
Each of these vendors does one thing well. However, they offer an incomplete solution for MLOps.
Domino Enterprise MLOps
Domino is unique in the marketplace. It supports all of the key MLOps capabilities in a single platform. Expert data scientists like Domino's flexibility and utility. Data science leaders like Domino project management, governance, and reproducibility. IT leaders like the way Domino consolidates data science tools on a single secure low-maintenance platform.