hello! Is it possible to backfill a metric for anomaly detection? i.e compute not just the latest date but also previous dates to see what the metric trend looks like 🤔

Hi @Kilian Tep, thanks for reaching out. If you'd fill the

timestamp_column

argument on your test, you'd be able to see the data's behavior over time.

Hi @Elon Gliksberg, thanks for the quick answer! I have actually specified the

timestamp_column

version: 2
models:
- name: enigma_scores
  description: Enigma scores
  config:
    elementary:
      timestamp_column: 'event_date_timestamp'
  columns:
  - name: avg_score
    description: Avg score
    tests:
      - elementary.column_anomalies:
          where_expression: turing_endpoint = "pj"
          column_anomalies:
            - sum
          time_bucket:
            period: day
            count: 1
          training_period:
            period: day
            count: 3

And I'm able to see the data's behavior from yesterday onwards -- when I first started running the test. My question is: can I plot the data prior to yesterday? Hope this clarifies.

The reason you're not seeing older data is because you've defined your

training_period

to only be 3 days. You can just remove this test argument in order to get the default 14 days, or set a larger value.

Thanks a lot, @Elon Gliksberg!!

My pleasure, glad I could help.