Hugging Face's logo

Datasets:
replikeit
/
rogii-oof-claude

Modalities:
Tabular
Text
Formats:
parquet
Size:
1M - 10M
Tags:
rogii
wellbore-geology
Libraries:
Datasets
pandas
Polars
License:
Dataset card Data Studio Files
xet
 Community
Dataset Viewer
Auto-converted to Parquet Duplicate
id
stringlengths
12
14
well
stringclasses
773 values
md
float64
10.9k
23k
truth
float64
10k
12.9k
recal_sel_lam045
float64
10k
12.9k
recal_stack_lam045
float64
10k
12.9k
recal_stack_lam030
float64
10k
12.9k
fold
int64
0
4
000d7d20_1442
000d7d20
12,909
11,747.38
11,747.399
11,748.108
11,748.171
2
000d7d20_1443
000d7d20
12,910
11,747.39
11,747.413
11,748.123
11,748.186
2
000d7d20_1444
000d7d20
12,911
11,747.4
11,747.425
11,748.138
11,748.201
2
000d7d20_1445
000d7d20
12,912
11,747.4
11,747.437
11,748.153
11,748.217
2
000d7d20_1446
000d7d20
12,913
11,747.41
11,747.451
11,748.169
11,748.233
2
000d7d20_1447
000d7d20
12,914
11,747.42
11,747.464
11,748.186
11,748.249
2
000d7d20_1448
000d7d20
12,915
11,747.43
11,747.47
11,748.197
11,748.261
2
000d7d20_1449
000d7d20
12,916
11,747.44
11,747.483
11,748.213
11,748.277
2
000d7d20_1450
000d7d20
12,917
11,747.45
11,747.496
11,748.229
11,748.292
2
000d7d20_1451
000d7d20
12,918
11,747.45
11,747.51
11,748.244
11,748.308
2
000d7d20_1452
000d7d20
12,919
11,747.46
11,747.537
11,748.253
11,748.317
2
000d7d20_1453
000d7d20
12,920
11,747.46
11,747.561
11,748.266
11,748.329
2
000d7d20_1454
000d7d20
12,921
11,747.47
11,747.572
11,748.278
11,748.341
2
000d7d20_1455
000d7d20
12,922
11,747.47
11,747.571
11,748.284
11,748.348
2
000d7d20_1456
000d7d20
12,923
11,747.48
11,747.576
11,748.293
11,748.357
2
000d7d20_1457
000d7d20
12,924
11,747.48
11,747.578
11,748.302
11,748.367
2
000d7d20_1458
000d7d20
12,925
11,747.49
11,747.569
11,748.305
11,748.371
2
000d7d20_1459
000d7d20
12,926
11,747.49
11,747.57
11,748.312
11,748.379
2
000d7d20_1460
000d7d20
12,927
11,747.5
11,747.566
11,748.314
11,748.382
2
000d7d20_1461
000d7d20
12,928
11,747.5
11,747.575
11,748.321
11,748.392
2
000d7d20_1462
000d7d20
12,929
11,747.5
11,747.574
11,748.324
11,748.396
2
000d7d20_1463
000d7d20
12,930
11,747.5
11,747.571
11,748.329
11,748.403
2
000d7d20_1464
000d7d20
12,931
11,747.51
11,747.59
11,748.342
11,748.416
2
000d7d20_1465
000d7d20
12,932
11,747.51
11,747.603
11,748.351
11,748.425
2
000d7d20_1466
000d7d20
12,933
11,747.51
11,747.616
11,748.363
11,748.437
2
000d7d20_1467
000d7d20
12,934
11,747.51
11,747.638
11,748.38
11,748.454
2
000d7d20_1468
000d7d20
12,935
11,747.51
11,747.652
11,748.393
11,748.466
2
000d7d20_1469
000d7d20
12,936
11,747.52
11,747.666
11,748.406
11,748.479
2
000d7d20_1470
000d7d20
12,937
11,747.52
11,747.681
11,748.418
11,748.491
2
000d7d20_1471
000d7d20
12,938
11,747.52
11,747.695
11,748.431
11,748.504
2
000d7d20_1472
000d7d20
12,939
11,747.52
11,747.711
11,748.443
11,748.515
2
000d7d20_1473
000d7d20
12,940
11,747.52
11,747.728
11,748.456
11,748.528
2
000d7d20_1474
000d7d20
12,941
11,747.51
11,747.749
11,748.468
11,748.539
2
000d7d20_1475
000d7d20
12,942
11,747.51
11,747.765
11,748.479
11,748.552
2
000d7d20_1476
000d7d20
12,943
11,747.51
11,747.78
11,748.49
11,748.564
2
000d7d20_1477
000d7d20
12,944
11,747.51
11,747.794
11,748.498
11,748.575
2
000d7d20_1478
000d7d20
12,945
11,747.51
11,747.808
11,748.507
11,748.586
2
000d7d20_1479
000d7d20
12,946
11,747.51
11,747.823
11,748.519
11,748.6
2
000d7d20_1480
000d7d20
12,947
11,747.51
11,747.837
11,748.531
11,748.616
2
000d7d20_1481
000d7d20
12,948
11,747.5
11,747.852
11,748.544
11,748.631
2
000d7d20_1482
000d7d20
12,949
11,747.5
11,747.867
11,748.558
11,748.648
2
000d7d20_1483
000d7d20
12,950
11,747.5
11,747.874
11,748.567
11,748.654
2
000d7d20_1484
000d7d20
12,951
11,747.49
11,747.889
11,748.58
11,748.666
2
000d7d20_1485
000d7d20
12,952
11,747.49
11,747.903
11,748.592
11,748.679
2
000d7d20_1486
000d7d20
12,953
11,747.48
11,747.91
11,748.595
11,748.681
2
000d7d20_1487
000d7d20
12,954
11,747.48
11,747.924
11,748.606
11,748.688
2
000d7d20_1488
000d7d20
12,955
11,747.48
11,747.937
11,748.62
11,748.697
2
000d7d20_1489
000d7d20
12,956
11,747.47
11,747.942
11,748.633
11,748.702
2
000d7d20_1490
000d7d20
12,957
11,747.47
11,747.954
11,748.646
11,748.708
2
000d7d20_1491
000d7d20
12,958
11,747.47
11,747.958
11,748.656
11,748.71
2
000d7d20_1492
000d7d20
12,959
11,747.47
11,747.97
11,748.67
11,748.713
2
000d7d20_1493
000d7d20
12,960
11,747.47
11,747.974
11,748.674
11,748.712
2
000d7d20_1494
000d7d20
12,961
11,747.47
11,747.985
11,748.68
11,748.715
2
000d7d20_1495
000d7d20
12,962
11,747.46
11,747.988
11,748.683
11,748.718
2
000d7d20_1496
000d7d20
12,963
11,747.46
11,747.99
11,748.684
11,748.722
2
000d7d20_1497
000d7d20
12,964
11,747.46
11,747.999
11,748.687
11,748.725
2
000d7d20_1498
000d7d20
12,965
11,747.46
11,748
11,748.692
11,748.726
2
000d7d20_1499
000d7d20
12,966
11,747.46
11,748.001
11,748.689
11,748.722
2
000d7d20_1500
000d7d20
12,967
11,747.46
11,748.005
11,748.686
11,748.721
2
000d7d20_1501
000d7d20
12,968
11,747.49
11,748.018
11,748.684
11,748.727
2
000d7d20_1502
000d7d20
12,969
11,747.51
11,748.025
11,748.688
11,748.73
2
000d7d20_1503
000d7d20
12,970
11,747.54
11,748.034
11,748.702
11,748.739
2
000d7d20_1504
000d7d20
12,971
11,747.56
11,748.046
11,748.721
11,748.754
2
000d7d20_1505
000d7d20
12,972
11,747.59
11,748.061
11,748.741
11,748.77
2
000d7d20_1506
000d7d20
12,973
11,747.61
11,748.08
11,748.758
11,748.786
2
000d7d20_1507
000d7d20
12,974
11,747.64
11,748.106
11,748.777
11,748.808
2
000d7d20_1508
000d7d20
12,975
11,747.66
11,748.119
11,748.789
11,748.822
2
000d7d20_1509
000d7d20
12,976
11,747.69
11,748.132
11,748.801
11,748.839
2
000d7d20_1510
000d7d20
12,977
11,747.71
11,748.141
11,748.821
11,748.855
2
000d7d20_1511
000d7d20
12,978
11,747.74
11,748.149
11,748.833
11,748.871
2
000d7d20_1512
000d7d20
12,979
11,747.76
11,748.154
11,748.846
11,748.885
2
000d7d20_1513
000d7d20
12,980
11,747.78
11,748.163
11,748.856
11,748.894
2
000d7d20_1514
000d7d20
12,981
11,747.81
11,748.176
11,748.884
11,748.906
2
000d7d20_1515
000d7d20
12,982
11,747.83
11,748.203
11,748.899
11,748.916
2
000d7d20_1516
000d7d20
12,983
11,747.86
11,748.236
11,748.919
11,748.927
2
000d7d20_1517
000d7d20
12,984
11,747.88
11,748.258
11,748.93
11,748.931
2
000d7d20_1518
000d7d20
12,985
11,747.9
11,748.276
11,748.947
11,748.935
2
000d7d20_1519
000d7d20
12,986
11,747.93
11,748.296
11,748.961
11,748.939
2
000d7d20_1520
000d7d20
12,987
11,747.95
11,748.318
11,748.965
11,748.938
2
000d7d20_1521
000d7d20
12,988
11,747.98
11,748.342
11,748.964
11,748.934
2
000d7d20_1522
000d7d20
12,989
11,748
11,748.358
11,748.95
11,748.923
2
000d7d20_1523
000d7d20
12,990
11,748.02
11,748.372
11,748.931
11,748.912
2
000d7d20_1524
000d7d20
12,991
11,748.05
11,748.389
11,748.913
11,748.901
2
000d7d20_1525
000d7d20
12,992
11,748.07
11,748.409
11,748.91
11,748.891
2
000d7d20_1526
000d7d20
12,993
11,748.09
11,748.432
11,748.91
11,748.886
2
000d7d20_1527
000d7d20
12,994
11,748.11
11,748.457
11,748.901
11,748.881
2
000d7d20_1528
000d7d20
12,995
11,748.14
11,748.477
11,748.898
11,748.88
2
000d7d20_1529
000d7d20
12,996
11,748.16
11,748.5
11,748.904
11,748.885
2
000d7d20_1530
000d7d20
12,997
11,748.18
11,748.516
11,748.919
11,748.89
2
000d7d20_1531
000d7d20
12,998
11,748.21
11,748.539
11,748.938
11,748.906
2
000d7d20_1532
000d7d20
12,999
11,748.23
11,748.561
11,748.96
11,748.92
2
000d7d20_1533
000d7d20
13,000
11,748.25
11,748.581
11,748.983
11,748.934
2
000d7d20_1534
000d7d20
13,001
11,748.27
11,748.6
11,749.009
11,748.95
2
000d7d20_1535
000d7d20
13,002
11,748.3
11,748.617
11,749.042
11,748.964
2
000d7d20_1536
000d7d20
13,003
11,748.32
11,748.635
11,749.058
11,748.972
2
000d7d20_1537
000d7d20
13,004
11,748.34
11,748.651
11,749.068
11,748.978
2
000d7d20_1538
000d7d20
13,005
11,748.36
11,748.668
11,749.082
11,748.982
2
000d7d20_1539
000d7d20
13,006
11,748.38
11,748.676
11,749.082
11,748.982
2
000d7d20_1540
000d7d20
13,007
11,748.41
11,748.691
11,749.082
11,748.976
2
000d7d20_1541
000d7d20
13,008
11,748.43
11,748.701
11,749.087
11,748.967
2
End of preview. Expand in Data Studio

ROGII — Claude's OOF train predictions (for teammate metric comparison)

Out-of-fold (honest, cross-fit) predictions on the 773 train wells, eval-zone rows only (where TVT_input is NaN — the rows that get scored). CV = GroupKFold(5) on WELLNAME (no well leakage): each well's prediction comes from a model trained on the OTHER 4 folds.

Engine: per-well affine typewell-GR recalibration (GR_well ≈ g·GR_tw(TVT)+b, Huber IRLS on the known zone, shrunk by λ) applied inside the particle filter emission (pf_recal_prod, 128 seeds × 500 particles), then a per-row zdyn residual LGBM stack (α=0.5 + U-smooth).

Columns

column meaning
id {WELLNAME}_{row_index} (Kaggle submission id format)
well, md well name, measured depth
truth TVT ground truth (you already have this — included for alignment)
recal_sel_lam045 recal-PF selector output, λ=0.45
recal_stack_lam045 recal-PF + zdyn stack, λ=0.45 — our ship/hedge, public LB 7.946
recal_stack_lam030 recal-PF + zdyn stack, λ=0.30 — public LB 8.071
fold GroupKFold(5) fold this well was held out in (0–4)

Pooled OOF RMSE (eval rows, all 773 wells)

  • recal_stack_lam045 = 9.258 (public LB 7.946)
  • recal_stack_lam030 = 9.831 (public LB 8.071)
  • recal_sel_lam045 = 9.708

Notes on λ (important)

The public LB is a BOWL in λ: 0.30→8.071, 0.45→7.946 (min), 0.60→8.499. The CV_A bowl min is higher (~0.60) but overshoots the LB — do NOT pick λ on CV alone; per-fold λ* is unstable {0.15,0.45,0.15,0.85,0.70}. λ=0.45 is our LB-best. These are OOF (train) numbers; multiply CV_A by ~0.79 for an LB estimate at a FIXED λ (the multiplier does NOT hold across λ).

Files: train_oof_claude.csv (270 MB) and train_oof_claude.parquet (91 MB, identical content).

(RU) Это мои out-of-fold предсказания на train (773 скважины, только eval-зона, GroupKFold(5) по WELLNAME — без утечки). Колонка recal_stack_lam045 — наш лучший объект (public LB 7.946). Truth включён для сверки. Метрики считай pooled RMSE по строкам. λ выбран по LB (bowl min на 0.45), не по CV.
Downloads last month
6
Number of rows:
3,783,989
Total file size:
362 MB