0.3579
Avg p-adic loss
1,476
Avg non-zero coefficients
10,000
Avg iterations
4,580
Total test samples
Method Description
This model implements Zubarev's p-adic polynomial regression method from arXiv:2503.23488. Key features:
- Initialization: Starts from UMLLR greedy coefficients
- Optimization: Simulated annealing with temperature-controlled random walk
- Perturbations: P-adic-aware (prefers powers of 79)
- Optional: Mahler polynomial basis for non-linear transforms
Cross-validation Results
| Fold | Mean p-adic loss | Non-zero coeffs | Iterations | Test samples |
|---|---|---|---|---|
| Fold 0 | 0.3895 | 2,594 | 10,000 | 884 |
| Fold 1 | 0.3570 | 712 | 10,000 | 959 |
| Fold 2 | 0.3457 | 718 | 10,000 | 918 |
| Fold 3 | 0.3337 | 741 | 10,000 | 918 |
| Fold 4 | 0.3638 | 2,613 | 10,000 | 901 |