Back to ULR overview · Back to main index
| Metric | Value |
|---|---|
| Test accuracy | 46.56% |
| Test F1 score | 0.5092 |
| Hierarchical loss | 0.89613142 |
| P-adic loss (total) | 738.75373481 |
| P-adic loss (mean) | 0.27340997 |
| Prime base | 79 |
| Number of tags (input features) | 4,981 |
| Non-zero parameters | 6,620 / 3,318,012 (99.8% sparse) |
| L1 regularization (C) | 1.0000 |
| Training samples | 10,857 |
| Test samples | 2,702 |
| Agreement | Count | Share | Cost per mistake | Total contribution |
|---|---|---|---|---|
| Exact match | 1,258 | 46.56% | 0.000000 | 0.000000 |
| p^6 | 2 | 0.07% | 0.000000 | 0.000000 |
| p^5 | 0 | 0.00% | 0.000000 | 0.000000 |
| p^4 | 52 | 1.92% | 0.000000 | 0.000001 |
| p^3 | 162 | 6.00% | 0.000002 | 0.000329 |
| p^2 | 278 | 10.29% | 0.000160 | 0.044544 |
| p^1 | 214 | 7.92% | 0.012658 | 2.708861 |
| p^0 | 736 | 27.24% | 1.000000 | 736.000000 |
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 (6,620) indicates how many coefficients the model actually uses.