Back to ULR overview · Back to main index
| Metric | Value |
|---|---|
| Test accuracy | 43.04% |
| Test F1 score | 0.4768 |
| Hierarchical loss | 0.89025442 |
| P-adic loss (total) | 1315.54406824 |
| P-adic loss (mean) | 0.28450347 |
| Prime base | 83 |
| Number of tags (input features) | 7,726 |
| Non-zero parameters | 10,772 / 7,665,184 (99.9% sparse) |
| L1 regularization (C) | 1.0000 |
| Training samples | 18,473 |
| Test samples | 4,624 |
| Agreement | Count | Share | Cost per mistake | Total contribution |
|---|---|---|---|---|
| Exact match | 1,990 | 43.04% | 0.000000 | 0.000000 |
| p^6 | 2 | 0.04% | 0.000000 | 0.000000 |
| p^5 | 10 | 0.22% | 0.000000 | 0.000000 |
| p^4 | 95 | 2.05% | 0.000000 | 0.000002 |
| p^3 | 255 | 5.51% | 0.000002 | 0.000446 |
| p^2 | 591 | 12.78% | 0.000145 | 0.085789 |
| p^1 | 370 | 8.00% | 0.012048 | 4.457831 |
| p^0 | 1,311 | 28.35% | 1.000000 | 1311.000000 |
L1 (Lasso) regularization promotes sparsity by driving many coefficients to exactly zero. This model uses ALL available tags (7,726) but L1 regularization selects which features are actually used. The number of non-zero parameters (10,772) indicates how many coefficients the model actually uses.