MLOps (Machine Learning Operations) is a combination of Machine Learning and DevOps principles. It uses DevOps concepts to manage the entire lifecycle of developing and deploying machine learning models. When developing a new ML-based service, the data science element is just the first step. Making the model operational in the live environment is where the complexity lies, and that’s what MLOps addresses.
With MLOps, the process of deploying a model is accelerated and streamlined, minimizing the level of DevOps effort, reducing time to market and improving model accuracy. MLOps seeks to increase automation and achieve a CI/CD approach to releasing production models.
MLOps builds and automates the entire machine learning lifecycle. These steps are:
- Data collection and processing
- Feature engineering
- Model training
- Model Deployment
- Model Testing
- Monitoring / drift detection
- Drift remediation
Governance is another important element of MLOps. Enterprises need tools and processes to ensure data quality and security, enable explainability, and allow appropriate data access for auditing purposes.