SPB Git

Repository Public Research preview

Commit

R03-EDIT-01/ADVE-01/ADVE-02/ADVE-03/ADVE-04/ADVE-16/ECON-01/ECON-06/EDIT-06/THEO-03/THEO-13/THEO-11: batch N1 metrics (M4BP48, 5 h)

OLS vs XGBoost (nested rounds) pooled out-of-fold R2 on the v2 design: random 0.646 vs 0.833; ZIP5-blocked 0.647 vs
0.815; ZIP3-blocked 0.635 vs 0.767; GroupKFold by state 0.580 vs 0.628; LOSO 0.573 vs 0.637; ten-state 0.541 vs 0.515
(XGB 1,000 rounds 0.425 reproduces v1.2; LightGBM nested 0.565). THEO-11: 1,856 held-out listings with |error| > 2
under the v1.1 OLS (not 204), carrying ~100% of the SSE; R2 = -1,955,902.5 reproduced.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Simon-Pierre Boucher committed · Nov 4, 2026 parent 6b7efa0
6 changed files +20,001 −0 Browse files
added results/geo/fold_membership.csv +70 −0
@@ -0,0 +1,70 @@
1 +design,fold,n_train,n_test,n_states_test,states_test,n_groups_test
2 +random,random,627102,156770,51,,
3 +size_matched,size_matched,348298,435574,51,,
4 +ten_state,ten_state,348319,435553,10,CA CO FL GA IL NC NY OH TX WA,10.0
5 +gkf_state,fold1,626868,157004,7,AK FL MT NH OR UT WY,7.0
6 +gkf_state,fold2,626685,157187,9,HI LA MD NE NJ RI TX VT WA,9.0
7 +gkf_state,fold3,627528,156344,11,AL AR AZ CA DE ID KS MI MN MO SD,11.0
8 +gkf_state,fold4,627142,156730,12,CO CT GA MS ND NV OH OK TN VA WI WV,12.0
9 +gkf_state,fold5,627265,156607,12,DC IA IL IN KY MA ME NC NM NY PA SC,12.0
10 +gkf_zip3,fold1,627097,156775,44,,177.0
11 +gkf_zip3,fold2,627095,156777,45,,177.0
12 +gkf_zip3,fold3,627105,156767,46,,178.0
13 +gkf_zip3,fold4,627097,156775,49,,177.0
14 +gkf_zip3,fold5,627094,156778,45,,177.0
15 +gkf_zip5,fold1,627097,156775,51,,5279.0
16 +gkf_zip5,fold2,627097,156775,51,,5281.0
17 +gkf_zip5,fold3,627098,156774,51,,5280.0
18 +gkf_zip5,fold4,627098,156774,51,,5280.0
19 +gkf_zip5,fold5,627098,156774,51,,5280.0
20 +loso,AK,783064,808,1,AK,1.0
21 +loso,AL,766953,16919,1,AL,1.0
22 +loso,AR,773207,10665,1,AR,1.0
23 +loso,AZ,760001,23871,1,AZ,1.0
24 +loso,CA,736222,47650,1,CA,1.0
25 +loso,CO,766392,17480,1,CO,1.0
26 +loso,CT,779745,4127,1,CT,1.0
27 +loso,DC,782463,1409,1,DC,1.0
28 +loso,DE,781874,1998,1,DE,1.0
29 +loso,FL,651686,132186,1,FL,1.0
30 +loso,GA,751884,31988,1,GA,1.0
31 +loso,HI,778859,5013,1,HI,1.0
32 +loso,IA,775607,8265,1,IA,1.0
33 +loso,ID,779340,4532,1,ID,1.0
34 +loso,IL,764702,19170,1,IL,1.0
35 +loso,IN,769534,14338,1,IN,1.0
36 +loso,KS,777898,5974,1,KS,1.0
37 +loso,KY,773235,10637,1,KY,1.0
38 +loso,LA,768938,14934,1,LA,1.0
39 +loso,MA,778571,5301,1,MA,1.0
40 +loso,MD,774206,9666,1,MD,1.0
41 +loso,ME,780606,3266,1,ME,1.0
42 +loso,MI,764682,19190,1,MI,1.0
43 +loso,MN,775162,8710,1,MN,1.0
44 +loso,MO,769088,14784,1,MO,1.0
45 +loso,MS,777099,6773,1,MS,1.0
46 +loso,MT,779913,3959,1,MT,1.0
47 +loso,NC,754627,29245,1,NC,1.0
48 +loso,ND,782533,1339,1,ND,1.0
49 +loso,NE,779947,3925,1,NE,1.0
50 +loso,NH,781583,2289,1,NH,1.0
51 +loso,NJ,776551,7321,1,NJ,1.0
52 +loso,NM,779168,4704,1,NM,1.0
53 +loso,NV,773112,10760,1,NV,1.0
54 +loso,NY,759049,24823,1,NY,1.0
55 +loso,OH,764464,19408,1,OH,1.0
56 +loso,OK,771626,12246,1,OK,1.0
57 +loso,OR,774387,9485,1,OR,1.0
58 +loso,PA,768745,15127,1,PA,1.0
59 +loso,RI,782832,1040,1,RI,1.0
60 +loso,SC,763550,20322,1,SC,1.0
61 +loso,SD,781821,2051,1,SD,1.0
62 +loso,TN,759275,24597,1,TN,1.0
63 +loso,TX,682203,101669,1,TX,1.0
64 +loso,UT,777248,6624,1,UT,1.0
65 +loso,VA,768153,15719,1,VA,1.0
66 +loso,VT,782187,1685,1,VT,1.0
67 +loso,WA,771938,11934,1,WA,1.0
68 +loso,WI,774758,9114,1,WI,1.0
69 +loso,WV,780693,3179,1,WV,1.0
70 +loso,WY,782219,1653,1,WY,1.0
added results/geo/metrics.csv +2974 −0
Large diff (2,974 lines) is not loaded by default to keep the page fast.Load diff
added results/geo/metrics_pooled.csv +167 −0
Large diff (167 lines) is not loaded by default to keep the page fast.Load diff
added results/geo/n_corrupt_heldout.csv +2 −0
Large diff (2 lines) is not loaded by default to keep the page fast.Load diff
added results/geo/rounds_selection.csv +6225 −0
Large diff (6,225 lines) is not loaded by default to keep the page fast.Load diff
added results/geo/trees_curves.csv +10563 −0
Large diff (10,563 lines) is not loaded by default to keep the page fast.Load diff

Diff truncated — file too large.

Menu

Theme