Donker All-Feature Tree Regression
Generated by scripts/run_donker_all_feature_tree_regression.py.
This tests p-adic exact through-points branch recovery against the two all-feature Donker tree targets: the direct core-witness tree and the exploratory all-label bridged tree.
| Target tree | Witnesses | Features | Nodes | Best model | Best loss | Best exact p-adic | Exact loss | Logistic loss | Exact - logistic |
|---|---|---|---|---|---|---|---|---|---|
all_labels_bridged |
41 | 219 | 39 | opaque_logistic_l1_C1 |
0.0572 | padic_exact_s3_k100 |
0.1063 | 0.0572 | +0.0490 |
core_all_sections |
8 | 139 | 6 | padic_exact_s2_k100 |
0.2860 | padic_exact_s2_k100 |
0.2860 | 0.3070 | -0.0210 |
The all-label target uses the same missing-as-zero numeric encoding as the existing p-adic experiments. Because many witnesses are absent from whole sections, this screen can exploit coverage patterns as well as reading-state patterns.
Files
- Combined summary:
outputs/tables/donker_all_feature_tree_regression_summary.csv - Per-target outputs:
outputs/tables/donker_all_feature_tree_regression/