Data format
What to send SymNexusForecast
The inputs the platform expects, and the labels it learns from.
Input
Input format
A long table with one row per series per time step: an ID, a date, and the value to forecast. Optional extra columns for things known in advance.
| store_id | date | sales | promo | price |
|---|---|---|---|---|
| store_01 | 2024-05-01 | 212.4 | 0 | 10.00 |
| store_01 | 2024-05-02 | 305.9 | 1 | 8.50 |
| store_02 | 2024-05-01 | 144.0 | 0 | 10.00 |
store_id,date,sales,promo,price
store_01,2024-05-01,212.4,0,10.00
store_01,2024-05-02,305.9,1,8.50
store_02,2024-05-01,144.0,0,10.00Labels
No separate labels are needed: the history itself is what the system learns from.
Tips for good results
- At least a few months of history per series; two years captures yearly seasonality.
- For future promotions or prices, also provide their planned values for the forecast period.
- Keep the time step regular (every day, every week…).