5. #azuresatpn
DevOps Workflow with ML Model lifecycle
• Application LifeCycle evolve in a extended Machine Learning Model
Lifecycle.
• ML model generation
• Training model
• Testing
• Evaluation
• Automatic Deployment
7. #azuresatpn
Evolve the pipeline
•
We need to add the following steps:
• Build/compile the ML model trainer app (Usually a console app)
• Run the process (console app) to train the ML.NET model and
generate the serialized model (.zip file).
• Run model’s tests (model quality validation)
• Deployment the model file into the actual end-user application code
(project structure)
8. #azuresatpn
Evolve the pipeline
•
And after we can return to our typical tasks:
• Build/compile the end-user app (such as an ASP.NET Core web app or
WebAPI service)
• Run app’s unit tests and integration tests
• Generate and publish the final pipeline artifact in Azure DevOps (or if
using containers, generate a Docker image and publish it into a
Docker Registry)
11. #azuresatpn
Improvements => the direction
1. Versioning datasets
2. Databases as training data
3. DevOps workflow Scenarios
4. ML Model Versioning
5. Integration with Azure ML and MLFlow