Challenge | Automation solution | Keep It Simple Solution | Complexity needed if … |
---|---|---|---|
Training-Serving Skew, where differences crop up between training and serving stacks | Feature Store | Transform Pattern | Features injected server-side |
Clients are interested in a capability that involves multiple steps, only one of which is the ML model | ML Pipelines | Stateless Serving Function | Cascade of ML models |
The incoming data changes characteristics | Continuous evaluation | Scheduled evaluation | Robust mitigation available |
The correct answer changes over time | Continuous training | Systematic triggers | Adaptive system |
Model fairness issues arise | Mumbo Jumbo | Manual sliced evaluations and other checks | Need to turn off |
Parts of the system aren’t changed when new code is checked in | CI/CD | From-scratch builds | Large repo, frequent model changes |
Need to be able to troubleshoot old results | Model registry, Data registry | Version control all deployed artifacts | Hundreds of models |
Last active
October 28, 2022 14:53
-
-
Save lakshmanok/ea06b04de6251d934a2b9518072850ed to your computer and use it in GitHub Desktop.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment