Benchmarks
Fixed tasks on standardized public data, with a split by well that never changes, so results from different papers can be compared. Cite the benchmark and the original data when you use them.
| Benchmark | Task | Split (train / val / test) | Best result |
|---|---|---|---|
OWD-DT-1 | Predict dt from gr, rhob, resistividad_profunda, profundidad | 82 / 19 / 17 | 12.16 |
OWD-DT-1: predict dt from other logs
Data: FORCE 2020 wells standardized by Open Well Data (CC BY 4.0). Inputs: gr, rhob, resistividad_profunda, profundidad. Fixed split by well: sha1(pozo) % 100: <70 train, <85 val, resto test. Wells: 82 train, 19 validation, 17 test.
| Model | RMSE (us/ft) | MAE (us/ft) | Test rows |
|---|---|---|---|
| Quadratic regression | 12.16 | 8.40 | 191,134 |
| Linear regression | 13.43 | 9.78 | 191,134 |
| Training mean | 29.27 | 25.69 | 191,134 |
Test wells: 15_9-17, 16_7-6, 25_10-10, 25_11-15, 25_11-19 S, 25_11-24, 25_6-1, 25_6-2, 25_6-3, 25_9-1, 31_2-9, 31_6-5, 34_10-33, 34_11-2 S, 34_7-20, 35_11-11, 7_1-2 S. To add a result, open an issue on GitHub with your code and the test metrics on this split.