search

cdfoundation / sig-mlops

CDF SIG MLOps
https://cd.foundation
Apache License 2.0
604 stars 69 forks source link

Update MLOpsRoadmap2020.md #27

Closed michaelneale closed 4 years ago

michaelneale commented 4 years ago

Added a challenge section on reinforcement learning and the "swearing problem"

michaelneale commented 4 years ago

What I am trying to capture is the need for MLOps to validate and prevent these cases (which I think is an MLOps problem). Either by gating the training data or at the least the model.

michaelneale commented 4 years ago

I think I will close this as the other PR combined the challenge and requirement together

michaelneale commented 4 years ago

Note: Need a challenge section added that online learning sits outside of the normal pipeline of train, test, deploy.. this PR could evolve to that. even if are retraining at runtime, perhaps there should be a feedback of the state of the model to the CI/CD environment for continuous verification (a kind of monitoring).

michaelneale commented 4 years ago

will open in a new PR