Back to ULR overview · Back to main index
| Metric | Value |
|---|---|
| Test accuracy | 46.77% |
| Test F1 score | 0.5104 |
| Hierarchical loss | 0.89708725 |
| P-adic loss (total) | 699.60932430 |
| P-adic loss (mean) | 0.26281342 |
| Prime base | 79 |
| Number of tags (input features) | 4,981 |
| Non-zero parameters | 6,750 / 3,322,994 (99.8% sparse) |
| L1 regularization (C) | 1.0000 |
| Training samples | 10,897 |
| Test samples | 2,662 |
| Agreement | Count | Share | Cost per mistake | Total contribution |
|---|---|---|---|---|
| Exact match | 1,245 | 46.77% | 0.000000 | 0.000000 |
| p^5 | 1 | 0.04% | 0.000000 | 0.000000 |
| p^4 | 51 | 1.92% | 0.000000 | 0.000001 |
| p^3 | 141 | 5.30% | 0.000002 | 0.000286 |
| p^2 | 325 | 12.21% | 0.000160 | 0.052075 |
| p^1 | 202 | 7.59% | 0.012658 | 2.556962 |
| p^0 | 697 | 26.18% | 1.000000 | 697.000000 |
L1 (Lasso) regularization promotes sparsity by driving many coefficients to exactly zero. This model uses ALL available tags (4,981) but L1 regularization selects which features are actually used. The number of non-zero parameters (6,750) indicates how many coefficients the model actually uses.