https://linen.dev logo
Join Slack
Powered by
Hi all! We have a table with events in it; I'd li...
# community-support
r
Hi all! We have a table with events in it; I'd like to run a dimensional anomaly test on this table with the event_name as the dimension. The events differ per-client though, with a few overlaps, so it's important to be able to group by client and test the dimension anomalies per-client. Generally if something is missing it's in one client, and not the others, due to the tagging setup being per-client and done by different CS folks. So there's an option for dimensions (sql expression) and an option for where (sql expression), but do either support group by? Or do I need to create a separate model with these pre-grouped in order to do this test? Thanks!
My second question is how to set this up so I get results. Just trying it out I get "Not enough data to calculate anomaly score" But I have several months of data. Why can't it calculate the score? What I'd like is for it to look at the past 30 days of data to determine whether the most recent data is missing events (or has a spike):
Copy code
- elementary.dimension_anomalies:
          timestamp_column: event_timestamp
          dimensions:
            - client_id
            - event_name
          training_period:
            period: day
            count: 30
          detection_period:
            period: day
            count: 3
          time_bucket:
            period: day
            count: 1
Looks like dimensional anomalies aren't working atm: https://github.com/elementary-data/elementary/issues/1170 I'll look into using BQ arima models instead, unless there's a possibility this will be fixed soon. (since it was logged in Sept, I'm guessing it's not a priority)
o
Hi @Renee Cooper, currently looking into it As for your first question, i think its a good idea to have a view with the normalized values, even regardless of elementary, for easier querying and ad-hoc operations for you
Open in Slack
PreviousNext
Powered by