Athanasius Ultrametric

Reports / further_branchpoint_analyses.md

Further Branch-point Analyses

Generated by scripts/run_further_branchpoint_analyses.py.

This report audits METHODOLOGICAL-GAPS.md against the current repo and runs reproducible follow-up checks from FURTHER-ANALYSES.md. Expensive nested p-adic checkpoints are reused where possible; the fixed-prediction permutation null is a calibration screen, while the bounded refit null reruns a smaller model set under permuted target assignments.

Methodological-gaps audit

gap current_status evidence remaining_work
1 selection-on-test / LOO used twice largely fixed computationally outputs/reports/branchpoint_nested_selection.md; paper/athanasius_ultrametric.tex Keep non-nested replay labelled as exploratory only.
2 singleton branch classes partly fixed outputs/tables/branchpoint_nested_singleton_summary.csv; outputs/tables/branchpoint_nested_per_witness.csv Promote singleton-aware table or explicit per-witness table into paper if needed.
3 seed SE vs generalisation partly fixed paper labels the one-SE replay as diagnostic and reports nested results separately Avoid using seed SE as statistical uncertainty outside the exploratory replay.
4 self-derived target tree circularity acknowledged, not fully fixed paper/athanasius_ultrametric.tex; outputs/reports/donker_ensemble_tree_regression.md Use the section-to-section transfer table generated by this script as the first direct test.
5 negative stability margins diagnosed, not resolved outputs/tables/prefix_vote_diagnostics.csv; outputs/tables/branchpoint_nested_padic_predictions.csv Use risk-coverage or choose larger/stabler operating points.
6 conditioning diagnostic unresolved still correct paper/tables/throughpoints_conditioning_summary.tex Either explain fragility with another variable or demote conditioning to appendix.
7 rational-to-Zp reduction partly fixed in code, underexplained in prose src/athanasius_ultrametric/padic_bruteforce.py; src/athanasius_ultrametric/ensemble_selection.py Add a worked denominator-divisible-by-5 example and state the loss cap convention.
8 baseline tuning asymmetry largely fixed outputs/tables/branchpoint_model_feature_prediction_summary.csv; paper/tables/branchpoint_model_feature_prediction_summary.tex The comparison is still tiny-n; do not rank methods strongly.
9 prefix-consensus / branch-risk naming mostly fixed paper title and branch-risk medoid proposition connect the names Keep one primary name in future manuscript edits.
10 orphan generated tables still correct in spirit 16 paper/tables/*.tex files are not input by paper/athanasius_ultrametric.tex Audit the remaining generated tables as promote/delete/appendix.

1. Is the p-adic medoid load-bearing?

The medoid ablation compares the saved branch-risk medoid with a recomputed medoid, recursive plurality over snapped branch digits, and flat majority over snapped branch labels.

dataset dataset_label method rows outer_folds repeats exact_branch_accuracy mean_padic_loss mean_medoid_gap mean_stability_margin agreement_with_saved
athanasius_epistles_data_bin_csv Binary branch_risk_medoid_recomputed 120 12 10 0.416667 0.223333 0.0723596 -0.473398 1
athanasius_epistles_data_bin_csv Binary branch_risk_medoid_saved 120 12 10 0.416667 0.223333 0.0723596 -0.473398 nan
athanasius_epistles_data_bin_csv Binary flat_majority_snapped 120 12 10 0.45 0.227333 0.0723596 -0.473398 0.816667
athanasius_epistles_data_bin_csv Binary recursive_plurality_snapped 120 12 10 0.316667 0.252667 0.0723596 -0.473398 0.508333
athanasius_epistles_data_multistate_csv Multistate branch_risk_medoid_recomputed 120 12 10 0.441667 0.170333 0.0732422 -0.592549 1
athanasius_epistles_data_multistate_csv Multistate branch_risk_medoid_saved 120 12 10 0.441667 0.170333 0.0732422 -0.592549 nan
athanasius_epistles_data_multistate_csv Multistate flat_majority_snapped 120 12 10 0.408333 0.194333 0.0732422 -0.592549 0.733333
athanasius_epistles_data_multistate_csv Multistate recursive_plurality_snapped 120 12 10 0.316667 0.286 0.0732422 -0.592549 0.566667

The linear baseline table compares branch-encoding ridge regression and the existing coordinate-descent p-adic linear model against the branch-point target.

dataset dataset_label method folds exact_branch_accuracy mean_padic_loss target_absent_folds
athanasius_epistles_data_bin_csv Binary padic_linear_coordinate_branch 12 0.166667 0.38 2
athanasius_epistles_data_bin_csv Binary real_ridge_branch_encoding 12 0.166667 0.606667 2
athanasius_epistles_data_multistate_csv Multistate padic_linear_coordinate_branch 12 0.166667 0.313333 2
athanasius_epistles_data_multistate_csv Multistate real_ridge_branch_encoding 12 0 0.44 2

Rational-denominator and branch-risk loss-cap diagnostics for the selected nested p-adic models:

dataset dataset_label folds selected_model_rows denominators_divisible_by_base folds_with_denominator_divisible_by_base true_losses_above_branch_risk_cap candidate_branch_losses_above_cap mean_candidate_branch_losses_at_unit_cap mean_raw_padic_loss_to_true mean_nested_padic_loss
athanasius_epistles_data_bin_csv Binary 12 376 0 0 0 0 4.33889 0.319816 0.223333
athanasius_epistles_data_multistate_csv Multistate 12 303 0 0 0 0 4.56651 0.361581 0.170333

2. Null calibration

Saved-prediction rescore null, which keeps the original selection fixed:

dataset_label comparison_group permutations actual_mean_padic_loss null_median_padic_loss null_p05_padic_loss null_p95_padic_loss loss_lower_tail_p_value actual_exact_branch_accuracy null_median_exact_branch_accuracy accuracy_upper_tail_p_value
Binary nested_decision_tree 200 0.16 0.47 0.286667 0.52 0.00497512 0.333333 0.166667 0.124378
Binary nested_logistic 200 0.0466667 0.406667 0.246667 0.573333 0.00497512 0.5 0.166667 0.0248756
Binary nested_padic_greedy 200 0.223333 0.490167 0.305717 0.58085 0.00497512 0.416667 0.15 0.00497512
Binary nested_random_forest 200 0.05 0.41 0.26 0.57 0.00497512 0.416667 0.166667 0.0298507
Multistate nested_decision_tree 200 0.126667 0.463333 0.209667 0.5035 0.00497512 0.5 0.166667 0.00995025
Multistate nested_logistic 200 0.113333 0.406667 0.193333 0.503333 0.00497512 0.5 0.166667 0.00995025
Multistate nested_padic_greedy 200 0.170333 0.485833 0.3161 0.550033 0.00497512 0.441667 0.145833 0.00497512
Multistate nested_random_forest 200 0.0466667 0.406667 0.24 0.573333 0.00497512 0.5 0.166667 0.00995025

Bounded full-refit target-permutation null. This refits dummy baselines, C=1 L2 opaque/path logistic models, and small exact p-adic ensembles; it is not the production K=200 nested p-adic procedure.

dataset_label model model_family permutations actual_mean_padic_loss null_median_padic_loss null_p05_padic_loss null_p95_padic_loss loss_lower_tail_p_value actual_exact_branch_accuracy null_median_exact_branch_accuracy accuracy_upper_tail_p_value
Binary opaque_logistic_l2_C1 opaque_logistic 50 0.06 0.45 0.225333 0.654667 0.0196078 0.5 0.0833333 0.0196078
Binary path_logistic_l2_C1 path_logistic 50 0.113333 0.44 0.179333 0.651833 0.0196078 0.5 0.0416667 0.0196078
Binary padic_exact_s1_k5 padic_exact_ensemble 50 0.286667 0.32 0.313333 0.413333 0.0196078 0.166667 0 0.490196
Binary dummy_training_padic_medoid dummy 50 0.32 0.32 0.32 0.32 0.764706 0 0 1
Binary padic_exact_s2_k5 padic_exact_ensemble 50 0.613333 0.356667 0.291333 0.460667 1 0 0.0416667 1
Binary dummy_root dummy 50 1 1 1 1 1 0 0 1
Multistate opaque_logistic_l2_C1 opaque_logistic 50 0.06 0.486667 0.283 0.647333 0.0196078 0.5 0 0.0392157
Multistate path_logistic_l2_C1 path_logistic 50 0.126667 0.48 0.248333 0.64 0.0196078 0.5 0 0.0392157
Multistate padic_exact_s1_k5 padic_exact_ensemble 50 0.153333 0.32 0.313333 0.4 0.0196078 0.166667 0.0833333 0.392157
Multistate padic_exact_s2_k5 padic_exact_ensemble 50 0.286667 0.341667 0.316667 0.487333 0.0196078 0.166667 0 0.176471
Multistate dummy_training_padic_medoid dummy 50 0.32 0.32 0.32 0.32 0.686275 0 0 1
Multistate dummy_root dummy 50 1 1 1 1 1 0 0 1

Paired fold comparison on non-singleton folds:

dataset dataset_label scope folds logistic_correct_folds padic_correct_folds logistic_only_correct padic_only_correct mcnemar_exact_p
athanasius_epistles_data_bin_csv Binary non_singleton_outer_folds_majority_padic_repeats 10 6 3 3 0 0.25
athanasius_epistles_data_multistate_csv Multistate non_singleton_outer_folds_majority_padic_repeats 10 6 5 2 1 1

3. Cross-section transfer

Existing ensemble-tree recoverability screen:

target_tree sections mean_best_loss mean_exact_loss mean_logistic_loss exact_beats_logistic exact_beats_dummy
direct_tree_distance 9 0.116286 0.169242 0.120178 1 6
exemplar_kernel 9 0.116597 0.157029 0.126262 4 5

Direct section-to-section transfer, with each target section tree projected onto the source/target witness overlap:

transfer_type rows target_sections source_sections mean_overlap_witnesses mean_best_loss median_best_loss mean_logistic_loss mean_exact_loss exact_best_rows logistic_best_rows dummy_best_rows
cross_section 72 9 9 14.6944 0.145145 0.120233 0.172422 0.199217 18 46 8
within_section 9 9 9 21.6667 0.0740292 0.0577256 0.0740292 0.169855 0 9 0

Best cross-section transfer rows by loss:

target_section source_section transfer_type overlap_witnesses source_features target_nodes padic_base best_model best_model_family best_mean_padic_loss best_dummy_loss best_logistic_loss best_exact_loss
donker_ath_acts_1_12 donker_ath_acts cross_section 24 44 16 11 path_logistic_l2_C1 path_logistic 0.00432893 0.503475 0.00432893 0.0888438
donker_ath_2cor_titus donker_ath_1cor cross_section 21 40 19 5 opaque_logistic_l2_C1 opaque_logistic 0.027445 0.58311 0.027445 0.160607
donker_ath_paulineepistles donker_ath_1cor cross_section 21 40 19 5 opaque_logistic_l2_C1 opaque_logistic 0.0386042 0.657143 0.0386042 0.168981
donker_ath_2cor_titus donker_ath_paulineepistles cross_section 21 163 19 5 path_logistic_l2_C1 path_logistic 0.0603301 0.58311 0.0603301 0.160212
donker_ath_heb donker_ath_1cor cross_section 21 40 16 5 path_logistic_l2_C1 path_logistic 0.0628736 0.53491 0.0628736 0.214496
donker_ath_1cor donker_ath_2cor_titus cross_section 21 69 19 5 opaque_logistic_l2_C1 opaque_logistic 0.0639238 0.612069 0.0639238 0.194636
donker_ath_1cor donker_ath_paulineepistles cross_section 21 163 19 5 path_logistic_l2_C1 path_logistic 0.0645029 0.612069 0.0645029 0.177036
donker_ath_heb donker_ath_2cor_titus cross_section 21 69 16 5 path_logistic_l1_C1 path_logistic 0.0646711 0.53491 0.0646711 0.212678
donker_ath_heb donker_ath_paulineepistles cross_section 21 163 16 5 opaque_logistic_l2_C1 opaque_logistic 0.0722487 0.53491 0.0722487 0.212678
donker_ath_paulineepistles donker_ath_2cor_titus cross_section 21 69 19 5 path_logistic_l2_C1 path_logistic 0.0729661 0.657143 0.0729661 0.205066
donker_ath_2cor_titus donker_ath_rom cross_section 21 19 19 5 path_logistic_l2_C1 path_logistic 0.0745399 0.58311 0.0745399 0.162812
donker_ath_acts donker_ath_acts_13_28 cross_section 24 25 21 5 opaque_logistic_l2_C1 opaque_logistic 0.0762805 0.539408 0.0762805 0.0781365

4. Sibling-labelling stress test

This is a post-fit branch-code relabelling stress test on saved selected raw predictions. It is not a full refit under random sibling orders.

dataset dataset_label relabelling permutations mean_exact_branch_accuracy p05_exact_branch_accuracy median_exact_branch_accuracy p95_exact_branch_accuracy mean_padic_loss_original_tree p05_padic_loss_original_tree median_padic_loss_original_tree p95_padic_loss_original_tree
athanasius_epistles_data_bin_csv Binary identity 1 0.416667 0.416667 0.416667 0.416667 0.223333 0.223333 0.223333 0.223333
athanasius_epistles_data_bin_csv Binary random_sibling_permutation 50 0.175667 0.00833333 0.225 0.416667 0.498413 0.223333 0.275 0.837333
athanasius_epistles_data_multistate_csv Multistate identity 1 0.441667 0.441667 0.441667 0.441667 0.170333 0.170333 0.170333 0.170333
athanasius_epistles_data_multistate_csv Multistate random_sibling_permutation 50 0.204167 0.05 0.233333 0.441667 0.475887 0.170333 0.236 0.846

5. Placeability diagnostics

The placeability score combines nested p-adic fold loss, medoid gap, stability margin, selected-feature homoplasy, and ultrametric-violation severity. The selected-feature homoplasy table counts only recurrent states touched by selected p-adic model features for the held-out witness.

dataset_label held_out selected_model_rows selected_feature_terms selected_recurrent_state_rate selected_homoplasy_excess_per_term active_selected_recurrent_state_rate active_selected_homoplasy_excess_per_active_term
Binary A 40 265 0.849057 1.49434 1 1.81111
Binary Ath 30 188 0.893617 1.52128 0 0
Binary B 38 245 0.853061 1.43673 1 1.97727
Binary C 30 192 0.880208 1.52604 1 1.90164
Binary D 32 209 0.866029 1.63636 0.92381 1.99048
Binary F 29 221 0.846154 1.47964 0.804124 1.3299
Binary G 30 231 0.87013 1.51082 0.866667 1.47619
Binary M1739 26 179 0.860335 1.59218 1 1.8871
Binary M223 28 185 0.864865 1.47568 0.969231 1.89231
Binary M2423 25 168 0.880952 1.47024 0.985294 1.83824
Binary P46 37 219 0.885845 1.44749 1 1.45312
Binary U1 31 187 0.86631 1.45455 1 1.73684
Multistate A 18 125 0.656 1.024 0.634615 0.894231
Multistate Ath 18 120 0.725 1.125 0.75 1.04348
Multistate B 26 167 0.820359 1.34731 0.833333 1.32639
Multistate C 36 217 0.764977 1.26728 0.778351 1.26804
Multistate D 35 156 0.698718 1.19872 0.676259 1.13669
Multistate F 26 168 0.613095 0.988095 0.613924 0.981013
Multistate G 29 165 0.606061 0.951515 0.596026 0.89404
Multistate M1739 24 150 0.773333 1.24 0.769784 1.20144
Multistate M223 27 187 0.721925 1.14439 0.709497 1.10056
Multistate M2423 21 147 0.809524 1.2585 0.801471 1.19853
Multistate P46 23 113 0.716814 1.16814 0.736842 1.15789
Multistate U1 20 139 0.705036 1.1223 0.697674 1.09302
dataset_label held_out exact_branch_accuracy mean_padic_loss mean_medoid_gap target_absent_from_training selected_homoplasy_excess_per_term active_selected_homoplasy_excess_per_active_term ultrametric_violation_severity_sum placeability_risk_score placeability_rank
Binary D 0 0.68 0.0300667 1 1.63636 1.99048 13.9099 6.51471 1
Binary M1739 0.3 0.188 0.0242 0 1.59218 1.8871 9.16661 3.60939 2
Binary C 0.5 0.308 0.01664 0 1.52604 1.90164 12.2148 3.3278 3
Binary A 0 0.392 0.133067 0 1.49434 1.81111 16.1596 1.12966 4
Binary P46 0 0.072 0.0315476 0 1.44749 1.45312 19.1916 1.0518 5
Binary M223 0.4 0.12 0.0313333 0 1.47568 1.89231 10.7504 0.555277 6
Binary B 0.5 0.324 0.0250222 0 1.43673 1.97727 10.1425 0.0465182 7
Binary U1 0.5 0.036 0.0179048 0 1.45455 1.73684 11.6815 -0.0339436 8
Binary M2423 0.8 0.12 0.117933 0 1.47024 1.83824 12.5232 -0.878427 9
Binary Ath 0 0.44 0.0975333 1 1.52128 0 6.83333 -3.89241 10
Binary G 1 0 0.166 0 1.51082 1.47619 9.74801 -5.04639 11
Binary F 1 0 0.177067 0 1.47964 1.3299 10.5241 -6.38399 12
Multistate D 0 0.68 0.0508 1 1.19872 1.13669 10.6667 4.72651 1
Multistate B 0.7 0.06 0.02504 0 1.34731 1.32639 7.4 4.1784 2
Multistate P46 0 0.168 0.0212 0 1.16814 1.15789 12.4 3.83375 3
Multistate M1739 0.4 0.232 0.0248 0 1.24 1.20144 7.4 3.47286 4
Multistate C 0.5 0.1 0.0333333 0 1.26728 1.26804 8.73333 1.37357 5
Multistate U1 0 0.152 0.0356 0 1.1223 1.09302 8.66667 1.33219 6
Multistate A 0.1 0.052 0.0329333 0 1.024 0.894231 9.46667 -0.0279539 7
Multistate M2423 0.9 0.02 0.123133 0 1.2585 1.19853 7.66667 -0.530964 8
Multistate M223 0.7 0.06 0.0823333 0 1.14439 1.10056 7.13333 -2.53522 9
Multistate Ath 0.1 0.42 0.193067 1 1.125 1.04348 6.6 -2.96442 10
Multistate F 1 0 0.126667 0 0.988095 0.981013 6.33333 -5.99899 11
Multistate G 0.9 0.1 0.13 0 0.951515 0.89404 6.33333 -6.85973 12

6. Risk-coverage

dataset dataset_label coverage_folds coverage_fraction exact_branch_accuracy mean_padic_loss mean_medoid_gap singleton_folds_covered covered_witnesses
athanasius_epistles_data_bin_csv Binary 1 0.0833333 1 0 0.177067 0 F
athanasius_epistles_data_bin_csv Binary 2 0.166667 1 0 0.171533 0 F; G
athanasius_epistles_data_bin_csv Binary 3 0.25 0.666667 0.130667 0.158711 0 F; G; A
athanasius_epistles_data_bin_csv Binary 4 0.333333 0.7 0.128 0.148517 0 F; G; A; M2423
athanasius_epistles_data_bin_csv Binary 5 0.416667 0.56 0.1904 0.13832 1 F; G; A; M2423; Ath
athanasius_epistles_data_bin_csv Binary 6 0.5 0.466667 0.170667 0.120525 1 F; G; A; M2423; Ath; P46
athanasius_epistles_data_bin_csv Binary 7 0.583333 0.457143 0.163429 0.107783 1 F; G; A; M2423; Ath; P46; M223
athanasius_epistles_data_bin_csv Binary 8 0.666667 0.4 0.228 0.0980685 2 F; G; A; M2423; Ath; P46; M223; D
athanasius_epistles_data_bin_csv Binary 9 0.75 0.411111 0.238667 0.0899522 2 F; G; A; M2423; Ath; P46; M223; D; B
athanasius_epistles_data_bin_csv Binary 10 0.833333 0.4 0.2336 0.083377 2 F; G; A; M2423; Ath; P46; M223; D; B; M1739
athanasius_epistles_data_bin_csv Binary 11 0.916667 0.409091 0.215636 0.077425 2 F; G; A; M2423; Ath; P46; M223; D; B; M1739; U1
athanasius_epistles_data_bin_csv Binary 12 1 0.416667 0.223333 0.0723596 2 F; G; A; M2423; Ath; P46; M223; D; B; M1739; U1; C
athanasius_epistles_data_multistate_csv Multistate 1 0.0833333 0.1 0.42 0.193067 1 Ath
athanasius_epistles_data_multistate_csv Multistate 2 0.166667 0.5 0.26 0.161533 1 Ath; G
athanasius_epistles_data_multistate_csv Multistate 3 0.25 0.666667 0.173333 0.149911 1 Ath; G; F
athanasius_epistles_data_multistate_csv Multistate 4 0.333333 0.725 0.135 0.143217 1 Ath; G; F; M2423
athanasius_epistles_data_multistate_csv Multistate 5 0.416667 0.72 0.12 0.13104 1 Ath; G; F; M2423; M223
athanasius_epistles_data_multistate_csv Multistate 6 0.5 0.6 0.213333 0.117667 2 Ath; G; F; M2423; M223; D
athanasius_epistles_data_multistate_csv Multistate 7 0.583333 0.514286 0.204571 0.105943 2 Ath; G; F; M2423; M223; D; U1
athanasius_epistles_data_multistate_csv Multistate 8 0.666667 0.5125 0.1915 0.0968667 2 Ath; G; F; M2423; M223; D; U1; C
athanasius_epistles_data_multistate_csv Multistate 9 0.75 0.466667 0.176 0.089763 2 Ath; G; F; M2423; M223; D; U1; C; A
athanasius_epistles_data_multistate_csv Multistate 10 0.833333 0.49 0.1644 0.0832907 2 Ath; G; F; M2423; M223; D; U1; C; A; B
athanasius_epistles_data_multistate_csv Multistate 11 0.916667 0.481818 0.170545 0.0779733 2 Ath; G; F; M2423; M223; D; U1; C; A; B; M1739
athanasius_epistles_data_multistate_csv Multistate 12 1 0.441667 0.170333 0.0732422 2 Ath; G; F; M2423; M223; D; U1; C; A; B; M1739; P46

7. Smaller robustness checks

Base sweep: saved branch predictions are rescored under alternative depth-discount bases. This does not refit the models; it asks whether the ranking changes when root-level errors are discounted more or less sharply.

dataset_label comparison_group evaluation_base rows exact_branch_accuracy prefix1_accuracy prefix2_accuracy mean_discounted_loss
Binary nested_logistic 5 12 0.5 1 0.833333 0.0466667
Binary nested_random_forest 5 12 0.416667 1 0.833333 0.05
Binary nested_decision_tree 5 12 0.333333 0.916667 0.583333 0.16
Binary nested_padic_greedy 5 120 0.416667 0.833333 0.583333 0.223333
Binary nested_logistic 7 12 0.5 1 0.833333 0.0306122
Binary nested_random_forest 7 12 0.416667 1 0.833333 0.0323129
Binary nested_decision_tree 7 12 0.333333 0.916667 0.583333 0.136054
Binary nested_padic_greedy 7 120 0.416667 0.833333 0.583333 0.205782
Binary nested_logistic 11 12 0.5 1 0.833333 0.0179063
Binary nested_random_forest 11 12 0.416667 1 0.833333 0.018595
Binary nested_decision_tree 11 12 0.333333 0.916667 0.583333 0.115702
Binary nested_padic_greedy 11 120 0.416667 0.833333 0.583333 0.190771
Multistate nested_random_forest 5 12 0.5 1 0.833333 0.0466667
Multistate nested_logistic 5 12 0.5 0.916667 0.833333 0.113333
Multistate nested_decision_tree 5 12 0.5 0.916667 0.75 0.126667
Multistate nested_padic_greedy 5 120 0.441667 0.883333 0.65 0.170333
Multistate nested_random_forest 7 12 0.5 1 0.833333 0.0306122
Multistate nested_logistic 7 12 0.5 0.916667 0.833333 0.102041
Multistate nested_decision_tree 7 12 0.5 0.916667 0.75 0.112245
Multistate nested_padic_greedy 7 120 0.441667 0.883333 0.65 0.153231
Multistate nested_random_forest 11 12 0.5 1 0.833333 0.0179063
Multistate nested_logistic 11 12 0.5 0.916667 0.833333 0.0936639
Multistate nested_decision_tree 11 12 0.5 0.916667 0.75 0.100551
Multistate nested_padic_greedy 11 120 0.441667 0.883333 0.65 0.138912

Approximate paired-design sample-size calculation from the observed non-singleton McNemar discordance. This is a planning diagnostic, not a confirmatory power analysis.

dataset_label observed_non_singleton_folds logistic_only_correct padic_only_correct discordant_fold_rate absolute_discordant_rate_gap approx_folds_for_80pct_power_alpha_0_05
Binary 10 3 0 0.3 0.3 27
Multistate 10 2 1 0.3 0.1 236

Feature redundancy among selected nested p-adic feature subsets, using existing full-correlation groups.

dataset_label selected_model_rows selected_feature_terms terms_in_correlated_groups terms_in_correlated_groups_share within_model_redundant_terms within_model_redundant_term_share mean_effective_group_share
Binary 376 2489 412 0.165528 0 0 1
Multistate 303 1854 260 0.140237 0 0 1