ML model — transparency
SpikeDesk retrains its prediction model on a rolling schedule. Every retrain runs a champion-challenger gate: the new model only goes live if it matches or beats the current one on a held-out validation window. If the challenger is weaker, the incumbent stays live and an alert goes out.
Promoted / rejected / failed
0 / 0 / 1
Recent retrains
| Finished | Result | New AUC | Prev AUC | Runtime | Notes |
|---|
| 2026-04-25 08:59 | timeout | — | — | 5400s | exceeded 5400s |
How to read this. "AUC" is the model's out-of-sample discrimination score on a held-out window; 0.50 is random, 1.00 is perfect. Anything in the 0.60–0.72 range on a rolling 3-year window is strong for F&O pattern detection. "Rejected" means the new model didn't beat the incumbent on validation, so we kept the old one — that's the gate doing its job, not a failure. Next eligible retrain: 2026-05-01 08:59 IST.
⚠
Informational analytics only. SpikeDesk is NOT SEBI-registered. This is not investment advice, research, or a buy/sell recommendation. Consult a SEBI-registered advisor. Full disclaimer →