Data format
What to send SymNexusPredict
The inputs the platform expects, and the labels it learns from.
Input
Input format
A table (CSV, Excel export or Parquet) with one row per record. Columns can be numbers, categories or dates; missing values are fine.
| customer_id | tenure_months | monthly_spend | support_tickets | contract | signup_date |
|---|---|---|---|---|---|
| 10482 | 3 | 120.00 | 4 | monthly | 2023-05-01 |
| 10483 | 60 | 49.99 | 0 | two_year | 2021-02-14 |
customer_id,tenure_months,monthly_spend,support_tickets,contract,signup_date
10482,3,120.00,4,monthly,2023-05-01
10483,60,49.99,0,two_year,2021-02-14For training
Label format
For training, one extra column holding the known outcome for past records (e.g. churned = 1 or 0, or the actual cost). New records to score don't need it.
| customer_id | churned |
|---|---|
| 10482 | 1 |
| 10483 | 0 |
Tips for good results
- Include only information known before the outcome happened.
- A few thousand historical rows is a good start.
- Keep IDs so predictions can be joined back to your systems.