Life-cycling stages

Deployment e ciclo di vita in MLOps

Nemanja Radojkovic

Senior Machine Learning Engineer

Machine Learning life cycles

 

1. ML project life cycle
2. ML application life cycle
3. ML model life cycle

A model is...

NOT something abstract and/or theoretical

A model IS

• a concrete, trained model
• ready to be put in use

Deployment

 

• Model object + deployment resources = model deployment package
• Putting our model in use = deployment

 

Model monitoring

 

• Is the model server running?
• Are the model inputs and outputs as expected?
• Also known as post-deployment monitoring

 

Decommissioning

 

 

Possible reasons

• Better model discovered
• More informative features created
• Modeled process changed, invalidating the existing model

 

=> Out with the old, in with the new

Model archiving

• Important in regulated industries.

  Why did your credit-risk model green-light this loan back in 2020?

 

• Reproducibility: ability to recreate, re-load, and use an old version of a model
  • Not a trivial requirement

 

Let's practice!

Deployment e ciclo di vita in MLOps