Validated on 2026-09-08 in an AWS Databricks workspace using serverless Jobs compute.
databricks bundle validate --strict -t dev --var="catalog=<target_catalog>": passed.- Deployment: passed.
- End-to-end lifecycle run: all serverless tasks passed.
- Independent batch run: configuration and scoring tasks passed.
- The deploy-time catalog/schema values reached both notebooks through task values; no notebook widgets or base parameters were used.
- 6 store series, 1,827 dates, and 10,962 regular store/date observations.
- 24 source gaps were forward-filled and retained as
was_imputed = true. - 0 remaining missing store-dates, duplicate keys, null targets, or negative targets.
| Rank | Model | SMAPE | WAPE | RMSE |
|---|---|---|---|---|
| 1 | AutoETS7 | 13.713 | 13.622 | 139,538.22 |
| 2 | AutoARIMA7 | 16.778 | 16.300 | 171,328.61 |
| 3 | SeasonalNaive7 | 18.210 | 18.407 | 189,109.30 |
| 4 | SeasonalNaive365 | 19.221 | 19.531 | 202,015.10 |
| 5 | Naive | 20.505 | 20.264 | 210,504.34 |
- UC model:
<target_catalog>.<target_schema>.store_revenue_forecaster - A new version was registered and resolved through the alias.
- Alias:
Champion - Selected algorithm:
AutoETS7
- Table:
<target_catalog>.<target_schema>.store_revenue_batch_forecasts - 180 rows: 6 stores × 30 future dates.
- Forecast range: 2003-01-03 through 2003-02-01.
- The resolved model version was recorded on every output row.
- Null predictions: 0.
Spark DataFrame cache() maps to PERSIST, which is unsupported on serverless Spark Connect. The
EDA notebook therefore recomputes its small summary query.