Machine Learning Model Deployment Strategies
Learn different deployment strategies for ML solutions to decide which one works best for your use case.
--
This article will explore different deployment strategies -Recreate, Canary, Blue-Green, and A/B testing. Advantages and disadvantages, and when to apply a particular deployment strategy to your Machine Learning solution.
Why do we need deployment strategies for ML Models?
A Machine learning model solves a business problem like identifying if a transaction is genuine or fraudulent; once the model has good performance metrics on unseen data as part of the validation step, it is ready to be integrated into a production system.
A deployment strategy is a way to update an existing ML Model deployed in production.
After deploying ML Model to production, monitor the model for any data drift or concept drift. When model decay is identified in production, then re-train the model with the new data or features. Machine learning models need to be deployed frequently and continuously to ensure expected performance; hence we need a deployment strategy.
Before you select a deployment strategy, answer the following questions
- What is an acceptable downtime SLA for your ML solution?
- How will you roll back the model in case the new model deployment has issues?
- Does your ML solution need backward compatibility?
- What will be the operation cost and overheads?
We will answer these questions as we explore each of the deployment strategies.
The different model deployment strategies are
- Recreate deployment
- Canary deployment
- Blue-Green deployment
- A/B testing deployment
Recreate Deployment
Recreate deployment will completely scale down the existing model before scaling up the new model version. Recreate strategy implies downtime for the ML solution as time is required to scale down the current model and…
