Exporting results
Forecast and exceptions CSV columns, and what each exception reason means.
Exports download the series currently visible on Results. Apply filters first if you want a subset. Files are CSV with a UTF-8 BOM.
Forecast export
Export forecast includes the run, the DFU, monthly dates, the point forecast, prediction ranges, and the selected model.
Columns:
run,unique_id,dimensions,ds,y_hat,y_hat_lo_80,y_hat_hi_80,y_hat_lo_95,y_hat_hi_95,best_model,metric,score,cold_start,cold_start_method
| Column | Meaning |
|---|---|
| run | Run version |
| unique_id | DFU identifier |
| dimensions | Hierarchy values |
| ds | Month (first day) |
| y_hat | Point forecast |
| y_hat_lo_80 / y_hat_hi_80 | 80% prediction range |
| y_hat_lo_95 / y_hat_hi_95 | 95% prediction range |
| best_model | Selected model |
| metric | Selection metric for the run |
| score | Selected model's score; empty for many cold starts |
| cold_start | Whether the DFU used a cold-start method |
| cold_start_method | Method used when cold_start is true |
Export dates use ds, even though the upload column is date. Range columns may be empty if that run did not persist bands for a row.
The forecast export does not include bias, a naive-baseline comparison, or scores for every candidate model.
Exceptions export
Export exceptions lists series that need attention.
Columns:
unique_id,dimensions,abc,best_model,metric,score,reasons
reasons can include one or more of:
| Reason | When it is set |
|---|---|
| cold_start | The series is a new / cold-start DFU |
| missing_score | The selected score is empty |
| short_history | 12 or fewer historical periods |
| intermittent | At least half of observations are zero |
| high_error | WMAPE or MAPE is at least 0.50, or MASE, MAE, or RMSE is at least 1 |
Open Model details on Results for the on-screen exceptions table without downloading.
Validation report
The Analysis step has a separate download for input errors and warnings (validation-{scenario}.csv). That file is about the upload, not the forecast output.