Back to ULR overview · Back to main index
| Metric | Value |
|---|---|
| Test accuracy | 55.79% |
| Test F1 score | 0.5932 |
| Hierarchical loss | 0.91377321 |
| P-adic loss (total) | 295.49949442 |
| P-adic loss (mean) | 0.30813295 |
| Prime base | 79 |
| Number of tags (input features) | 6,936 |
| Non-zero parameters | 2,651 / 2,407,139 (99.9% sparse) |
| L1 regularization (C) | 1.0000 |
| Training samples | 3,621 |
| Test samples | 959 |
| Agreement | Count | Share | Cost per mistake | Total contribution |
|---|---|---|---|---|
| Exact match | 539 | 56.20% | 0.000000 | 0.000000 |
| p^5 | 5 | 0.52% | 0.000000 | 0.000000 |
| p^4 | 18 | 1.88% | 0.000000 | 0.000000 |
| p^3 | 27 | 2.82% | 0.000002 | 0.000055 |
| p^2 | 36 | 3.75% | 0.000160 | 0.005768 |
| p^1 | 39 | 4.07% | 0.012658 | 0.493671 |
| p^0 | 295 | 30.76% | 1.000000 | 295.000000 |
L1 (Lasso) regularization promotes sparsity by driving many coefficients to exactly zero. This model uses ALL available tags (6,936) but L1 regularization selects which features are actually used. The number of non-zero parameters (2,651) indicates how many coefficients the model actually uses.