Dataset Viewer
Auto-converted to Parquet Duplicate
Search is not available for this dataset
id1
int64
25
850B
id2
int64
24
850B
predict
float64
0
1
25
506,806,211,572
0.145115
253
816,043,813,006
0.195594
317
790,274,076,243
0.661048
413
841,813,713,588
0.141281
435
3,459
0.309024
509
515,396,104,691
0.48932
617
747,324,444,834
0.447943
755
68,719,519,497
0.365683
1,062
317,827,706,265
0.439988
1,118
283,467,857,691
0.106691
1,120
309,237,699,720
0.63341
1,167
154,618,878,913
0.086323
1,376
627,065,262,726
0.885631
1,529
712,964,606,816
0.185947
1,553
712,964,682,144
0.212069
1,832
163,208,777,214
0.146087
1,860
695,784,720,364
0.865224
1,868
137,438,983,561
0.894789
1,899
592,705,592,965
0.108189
1,913
68,719,516,620
0.701713
1,913
223,338,343,705
0.701713
1,916
146,028,977,882
0.03622
1,917
68,719,597,305
0.481576
2,057
712,964,708,361
0.065605
2,215
377,957,259,749
0.221351
2,223
34,359,869,264
0.631594
2,256
498,216,285,727
0.198064
2,572
438,086,797,218
0.239349
2,835
223,338,342,569
0.336331
2,994
369,367,228,908
0.284576
3,093
231,928,264,691
0.152032
3,185
163,208,852,952
0.832237
3,259
77,309,517,670
0.038466
3,336
790,274,128,614
0.655777
3,370
446,676,670,716
0.066569
3,485
188,978,603,213
0.33112
3,680
644,245,102,679
0.73412
3,699
687,194,829,255
0.721743
3,708
558,345,751,189
0.810479
3,713
103,079,244,394
0.306527
3,770
730,144,548,491
0.731059
3,860
386,547,208,273
0.824462
4,028
77,309,508,125
0.086323
4,093
446,676,601,510
0.216012
4,189
523,986,087,063
0.534612
4,248
94,489,310,542
0.235108
4,284
695,784,835,334
0.20947
4,306
266,288,005,171
0.160266
4,357
489,626,387,895
0.620162
4,399
523,986,014,847
0.919643
4,437
412,316,879,807
0.350421
4,451
824,633,769,395
0.796925
4,478
51,539,681,976
0.68763
4,478
154,618,928,644
0.661923
4,478
429,496,743,143
0.555676
4,478
429,496,793,567
0.602062
4,790
180,388,685,484
0.287768
4,858
481,036,477,311
0.044019
4,941
146,029,036,269
0.115961
5,154
489,626,395,632
0.165593
5,190
309,237,723,427
0.279828
5,221
678,604,881,028
0.33112
5,246
68,719,480,164
0.08389
5,443
481,036,499,604
0.119203
5,461
455,266,612,318
0.087564
5,493
68,719,537,851
0.446736
5,496
360,777,266,005
0.180107
5,502
678,604,904,877
0.125065
5,524
584,115,661,410
0.426799
5,603
309,237,774,315
0.35488
5,666
592,705,505,301
0.144149
5,705
120,259,230,308
0.183594
5,772
833,223,681,971
0.341583
5,788
395,137,085,742
0.690976
5,829
8,589,995,999
0.399343
5,829
360,777,261,511
0.578448
5,829
670,014,934,181
0.593138
5,831
420,906,812,190
0.948917
5,990
566,935,821,643
0.338952
6,017
8,589,981,788
0.949669
6,029
695,784,797,807
0.168857
6,069
618,475,435,199
0.454714
6,128
558,345,813,858
0.738669
6,150
283,467,848,205
0.606732
6,226
249,108,217,546
0.345105
6,254
773,094,166,440
0.218669
6,281
635,655,293,041
0.108189
6,287
386,547,173,244
0.823328
6,319
103,079,317,711
0.933431
6,339
472,446,442,107
0.294215
6,397
438,086,698,856
0.140336
6,400
146,029,051,965
0.178956
6,530
395,137,027,803
0.288569
6,549
386,547,077,921
0.861538
6,563
17,179,947,298
0.08882
6,590
103,079,293,466
0.927363
6,662
781,684,136,051
0.261331
6,683
446,676,670,676
0.224055
6,696
463,856,527,392
0.151028
6,730
704,374,758,998
0.176672
End of preview. Expand in Data Studio

E-CUP 2026 retrospective evaluation proxy v2

Private team dataset for reproducing the checksum-bound retrospective contest transfer diagnostic.

Primary panel

  • file: soft_sample_pairs.parquet
  • immutable revision: 183601763fd7f4d1695325b315ef3c7cc98e67c1
  • immutable download: https://huggingface.co/datasets/mariklolik/ecup-2026-matching-eval-proxy-v2/resolve/183601763fd7f4d1695325b315ef3c7cc98e67c1/soft_sample_pairs.parquet
  • rows: 250000
  • SHA-256: f12abf3babed51f63439c9db023474aa32f682a92c2b97b2948b2fdee43df57b
  • population: 11187780
  • seed: 20260822
  • sampling: sorted numpy.random.default_rng(seed).choice(population, 250000, replace=False)
  • category: items.category joined by id1

Evaluator

The evaluator commit predates this private Hub publication, so its --dataset-info output still records the former distribution block. The immutable URL above supersedes only that distribution status; scoring semantics are unchanged.

Download and evaluate:

hf download mariklolik/ecup-2026-matching-eval-proxy-v2 \
  soft_sample_pairs.parquet \
  --repo-type dataset \
  --local-dir evaluation-proxy

python contest_score_proxy.py predictions.csv \
  --pairs evaluation-proxy/soft_sample_pairs.parquet \
  --details

Prediction CSV must contain exactly id1,id2,predict, use the panel's exact row order, contain no duplicate/null/non-finite values, and retain integer IDs.

The primary scalar applies the official category-wise unweighted sklearn.average_precision_score semantics to a row-level two-shift prior model. Fractional LLM votes are explicitly downweighted. Tie handling matches sklearn exactly.

Applicability boundary

This is retrospective_only evidence. It reproduced the ordering of five historical submissions with MAE 0.0015250420378983165 and maximum absolute error 0.002407512449966648, but it has no prospective validation and does not identify hidden Public truth. Use it as transfer evidence beside component-disjoint human OOF, never as a standalone promotion gate.

The panel and historical outputs are derived from private competition data and remain private team assets.

Downloads last month
32