ML Deployment

Created: 2023-02-13 15:39
#note

Embedded software:

• functions in a program;
• SDK;
• Python package;
• Pipeline Lambda;
• Bundled solution.

ML as a Service:

• Batch ML job;
• Request/response API;
• Request/response API;
• Stream processing job.

We generally need to convert a Notebook to Software and then we need to do regression testing (test if performance are not worse than before and if pre/post-processing get the same results). See Deployment best practices, Rollout Strategies, and Challenges for ML Deployment for detailed guidance on these aspects.

Tags

#mlops #ml #deployment