Does applying custom rules on machine learning jobs altert the ML model?

[ElasticLiver (Dense Man)]

Hi, I was wondering if applying custom rules on machine learning jobs altert the ML model? or the ML model stays the same, just those anomalies are not shown?

[richcollier]

It can do either/or/both.

Skip result = don't create an anomaly if the condition is met
Skip model update = don't alter the model if the condition is met

[ElasticLiver (Dense Man)]

Thanks Rich, So if in a metric I have values from 1 to 100 and I choose to skip the model update and skip values over and under, in a way ML will "know" that my metric moves in that range and it will give me better anomaly results?

[richcollier]

not quite understanding your question... If your metric is in the range of 1 to 100 (always?) then it doesn't make sense to skip over/under if they never go over/under. Or maybe, you're trying to protect yourself from badly reported data?

[tomyui (Tom)]

How about:

If the metric NORMALLY is in the range of 1 to 100.
But sometimes it would go under/over, which is an anomaly.

Does skipping model update give me better result?
And should we skip model update in case of anomaly? is it a good practice?

[richcollier]

Elastic ML already tries to not let anomalous values overly affect the model - so trying to manage it on your own with custom rules, while possible, is likely not really worth the effort.

However, with that said, if you ever get into a position where you feel like the model is "wrecked" by some unruly data, you can revert the model to a previous snapshot (to a time before your unruly input data): Model snapshots | Machine Learning in the Elastic Stack [7.14] | Elastic