Datasets:

kpshinnik
/

repro-spa-search-ad-bundle

ArXiv:

Dataset Preview

Split (1)

train

The full dataset viewer is not available (click to read why). Only showing a preview of the rows.

The dataset generation failed because of a cast error

Error code:   DatasetGenerationCastError
Exception:    DatasetGenerationCastError
Message:      An error occurred while generating the dataset

All the data files must have the same columns, but at some point there are 2 new columns ({'regret', 'T'}) and 6 missing columns ({'linucb_std', 'causal_mean', 't', 'causal_std', 'linucb_mean', 'sqrt_ref'}).

This happened while the csv dataset builder was generating data using

hf://datasets/kpshinnik/repro-spa-search-ad-bundle/outputs/claim1_scaling_T.csv (at revision 62dd9a93ec700302ff4c160404126167e52896ac), ['hf://datasets/kpshinnik/repro-spa-search-ad-bundle@62dd9a93ec700302ff4c160404126167e52896ac/outputs/claim1_figure2.csv', 'hf://datasets/kpshinnik/repro-spa-search-ad-bundle@62dd9a93ec700302ff4c160404126167e52896ac/outputs/claim1_scaling_T.csv', 'hf://datasets/kpshinnik/repro-spa-search-ad-bundle@62dd9a93ec700302ff4c160404126167e52896ac/outputs/claim1_scaling_d.csv', 'hf://datasets/kpshinnik/repro-spa-search-ad-bundle@62dd9a93ec700302ff4c160404126167e52896ac/outputs/claim2_knowng_T.csv', 'hf://datasets/kpshinnik/repro-spa-search-ad-bundle@62dd9a93ec700302ff4c160404126167e52896ac/outputs/claim3_spa_vs_fpa.csv', 'hf://datasets/kpshinnik/repro-spa-search-ad-bundle@62dd9a93ec700302ff4c160404126167e52896ac/outputs/claim5_error_vs_n.csv', 'hf://datasets/kpshinnik/repro-spa-search-ad-bundle@62dd9a93ec700302ff4c160404126167e52896ac/outputs/claim5_interval_counts.csv', 'hf://datasets/kpshinnik/repro-spa-search-ad-bundle@62dd9a93ec700302ff4c160404126167e52896ac/outputs/claim5_recovery_truncnorm.csv', 'hf://datasets/kpshinnik/repro-spa-search-ad-bundle@62dd9a93ec700302ff4c160404126167e52896ac/outputs/claim5_recovery_uniform.csv']

Please either edit the data files to have matching columns, or separate them into different configurations (see docs at https://hf.co/docs/hub/datasets-manual-configuration#multiple-configurations)
Traceback:    Traceback (most recent call last):
                File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1837, in _prepare_split_single
                  writer.write_table(table)
                  ~~~~~~~~~~~~~~~~~~^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/arrow_writer.py", line 765, in write_table
                  self._write_table(pa_table, writer_batch_size=writer_batch_size)
                  ~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/arrow_writer.py", line 773, in _write_table
                  pa_table = table_cast(pa_table, self._schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2369, in table_cast
                  return cast_table_to_schema(table, schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2297, in cast_table_to_schema
                  raise CastError(
                  ...<3 lines>...
                  )
              datasets.table.CastError: Couldn't cast
              T: int64
              regret: double
              -- schema metadata --
              pandas: '{"index_columns": [{"kind": "range", "name": null, "start": 0, "' + 489
              to
              {'t': Value('int64'), 'causal_mean': Value('float64'), 'causal_std': Value('float64'), 'linucb_mean': Value('float64'), 'linucb_std': Value('float64'), 'sqrt_ref': Value('float64')}
              because column names don't match
              
              During handling of the above exception, another exception occurred:
              
              Traceback (most recent call last):
                File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 1369, in compute_config_parquet_and_info_response
                  parquet_operations, partial, estimated_dataset_info = stream_convert_to_parquet(
                                                                        ~~~~~~~~~~~~~~~~~~~~~~~~~^
                      builder, max_dataset_size_bytes=max_dataset_size_bytes
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                  )
                  ^
                File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 948, in stream_convert_to_parquet
                  builder._prepare_split(split_generator=splits_generators[split], file_format="parquet")
                  ~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1683, in _prepare_split
                  for job_id, done, content in self._prepare_split_single(
                                               ~~~~~~~~~~~~~~~~~~~~~~~~~~^
                      gen_kwargs=gen_kwargs, job_id=job_id, **_prepare_split_args
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                  ):
                  ^
                File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1839, in _prepare_split_single
                  raise DatasetGenerationCastError.from_cast_error(
                  ...<4 lines>...
                  )
              datasets.exceptions.DatasetGenerationCastError: An error occurred while generating the dataset
              
              All the data files must have the same columns, but at some point there are 2 new columns ({'regret', 'T'}) and 6 missing columns ({'linucb_std', 'causal_mean', 't', 'causal_std', 'linucb_mean', 'sqrt_ref'}).
              
              This happened while the csv dataset builder was generating data using
              
              hf://datasets/kpshinnik/repro-spa-search-ad-bundle/outputs/claim1_scaling_T.csv (at revision 62dd9a93ec700302ff4c160404126167e52896ac), ['hf://datasets/kpshinnik/repro-spa-search-ad-bundle@62dd9a93ec700302ff4c160404126167e52896ac/outputs/claim1_figure2.csv', 'hf://datasets/kpshinnik/repro-spa-search-ad-bundle@62dd9a93ec700302ff4c160404126167e52896ac/outputs/claim1_scaling_T.csv', 'hf://datasets/kpshinnik/repro-spa-search-ad-bundle@62dd9a93ec700302ff4c160404126167e52896ac/outputs/claim1_scaling_d.csv', 'hf://datasets/kpshinnik/repro-spa-search-ad-bundle@62dd9a93ec700302ff4c160404126167e52896ac/outputs/claim2_knowng_T.csv', 'hf://datasets/kpshinnik/repro-spa-search-ad-bundle@62dd9a93ec700302ff4c160404126167e52896ac/outputs/claim3_spa_vs_fpa.csv', 'hf://datasets/kpshinnik/repro-spa-search-ad-bundle@62dd9a93ec700302ff4c160404126167e52896ac/outputs/claim5_error_vs_n.csv', 'hf://datasets/kpshinnik/repro-spa-search-ad-bundle@62dd9a93ec700302ff4c160404126167e52896ac/outputs/claim5_interval_counts.csv', 'hf://datasets/kpshinnik/repro-spa-search-ad-bundle@62dd9a93ec700302ff4c160404126167e52896ac/outputs/claim5_recovery_truncnorm.csv', 'hf://datasets/kpshinnik/repro-spa-search-ad-bundle@62dd9a93ec700302ff4c160404126167e52896ac/outputs/claim5_recovery_uniform.csv']
              
              Please either edit the data files to have matching columns, or separate them into different configurations (see docs at https://hf.co/docs/hub/datasets-manual-configuration#multiple-configurations)

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

t int64	causal_mean float64	causal_std float64	linucb_mean float64	linucb_std float64	sqrt_ref float64
1	0.531352	0.163183	0.044478	0.03325	5.98084
126	63.5165	1.74834	19.8117	0.464365	67.1347
251	125.072	2.89644	37.024	0.601525	94.7543
376	188.851	3.18292	52.729	0.691611	115.973
502	252.541	3.47398	67.7002	0.703536	134.003
627	315.585	3.83058	82.0095	0.775818	149.76
752	379.69	5.17367	95.88	0.818527	164.01
878	443.16	5.67181	109.556	0.814577	177.219
1,003	505.952	5.68539	122.963	0.843607	189.414
1,128	569.636	6.01136	136.125	0.824971	200.871
1,254	633.013	5.49526	149.266	0.887581	211.793
1,379	695.883	5.66347	162.101	0.932533	222.098
1,504	758.603	6.95939	174.811	0.933677	231.946
1,630	822.005	6.67546	187.482	0.91037	241.466
1,755	885.298	7.59828	199.908	0.870539	250.554
1,880	948.402	7.28056	212.295	0.882122	259.323
2,005	1,012.37	7.09346	224.51	0.944406	267.805
2,131	1,078.01	6.89135	236.757	0.945838	276.092
2,256	1,141.88	7.09524	248.856	0.933035	284.074
2,381	1,205.24	7.1945	260.926	0.898652	291.838
2,507	1,238.46	7.57668	272.972	0.930085	299.46
2,632	1,250.1	7.94134	284.846	1.02501	306.835
2,757	1,257.53	8.03232	296.657	1.08488	314.037
2,883	1,263.05	8.08588	308.581	1.09346	321.133
3,008	1,267.16	8.0053	320.219	1.13721	328.02
3,133	1,270.66	8.11218	331.916	1.16607	334.767
3,259	1,273.53	8.20352	343.732	1.11469	341.432
3,384	1,275.88	8.34101	355.292	1.20326	347.918
3,509	1,278	8.54245	366.828	1.2621	354.286
3,635	1,280.04	8.79252	378.518	1.25136	360.59
3,760	1,281.89	8.9792	389.986	1.22003	366.738
3,885	1,283.61	9.21924	401.536	1.15906	372.784
4,010	1,285.14	9.47582	413.001	1.20221	378.734
4,136	1,286.6	9.70419	424.572	1.2004	384.638
4,261	1,287.92	9.94549	435.978	1.19992	390.407
4,386	1,289.21	10.3387	447.339	1.2288	396.092
4,512	1,290.38	10.6268	458.754	1.26559	401.741
4,637	1,291.52	10.9363	469.993	1.21119	407.268
4,762	1,292.55	11.1524	481.253	1.24074	412.721
4,888	1,293.56	11.455	492.586	1.22979	418.146
5,013	1,294.46	11.652	503.816	1.20138	423.459
5,138	1,295.34	11.8783	515.026	1.21455	428.706
5,264	1,296.25	12.1044	526.323	1.21701	433.93
5,389	1,297.11	12.3616	537.518	1.21096	439.052
5,514	1,297.91	12.6392	548.629	1.3058	444.115
5,639	1,298.69	12.829	559.75	1.32506	449.121
5,765	1,299.49	13.087	571.025	1.35576	454.111
5,890	1,300.21	13.3143	582.082	1.40049	459.007
6,015	1,300.84	13.4641	593.024	1.44735	463.852
6,141	1,301.49	13.5848	604.166	1.45626	468.686
6,266	1,302.13	13.7593	615.182	1.48466	473.432
6,391	1,302.72	13.8897	626.269	1.51593	478.131
6,517	1,303.3	14.0087	637.392	1.55333	482.821
6,642	1,303.89	14.1597	648.379	1.53054	487.429
6,767	1,304.46	14.3188	659.355	1.58986	491.994
6,893	1,305.01	14.466	670.413	1.58748	496.554
7,018	1,305.55	14.6232	681.386	1.58828	501.036
7,143	1,306.01	14.7625	692.252	1.66355	505.478
7,269	1,306.51	14.8941	703.284	1.64041	509.917
7,394	1,306.96	15.0236	714.171	1.69705	514.283
7,519	1,307.41	15.1839	724.982	1.70899	518.611
7,644	1,307.86	15.2854	735.839	1.74853	522.905
7,770	1,308.27	15.4031	746.684	1.71615	527.197
7,895	1,308.69	15.5398	757.53	1.72706	531.42
8,020	1,309.12	15.664	768.406	1.7896	535.611
8,146	1,309.52	15.7557	779.316	1.80388	539.802
8,271	1,309.95	15.8859	790.204	1.82609	543.928
8,396	1,310.33	15.9902	801.017	1.82609	548.022
8,522	1,310.69	16.0768	811.92	1.77039	552.119
8,647	1,311.09	16.2235	822.729	1.82695	556.154
8,772	1,311.46	16.3329	833.483	1.79197	560.159
8,898	1,311.81	16.4033	844.301	1.81903	564.168
9,023	1,312.16	16.5101	855.061	1.84391	568.117
9,148	1,312.52	16.6259	865.866	1.82147	572.038
9,273	1,312.82	16.7172	876.511	1.8471	575.933
9,399	1,313.17	16.8201	887.301	1.88165	579.833
9,524	1,313.46	16.8924	897.956	1.76489	583.676
9,649	1,313.76	16.9624	908.657	1.73246	587.494
9,775	1,314.04	17.0456	919.409	1.71499	591.317
9,900	1,314.32	17.1563	930.071	1.81374	595.086
10,025	1,314.6	17.2105	940.686	1.77763	598.831
10,151	1,314.88	17.2929	951.475	1.71994	602.582
10,276	1,315.12	17.3475	962.087	1.7535	606.281
10,401	1,315.37	17.4079	972.848	1.72544	609.958
10,527	1,315.63	17.4877	983.523	1.71526	613.641
10,652	1,315.88	17.5583	994.187	1.70923	617.274
10,777	1,316.12	17.6289	1,004.78	1.66394	620.885
10,903	1,316.36	17.7008	1,015.51	1.72935	624.504
11,028	1,316.6	17.7676	1,026.13	1.73774	628.073
11,153	1,316.84	17.8439	1,036.74	1.76256	631.623
11,278	1,317.06	17.9027	1,047.38	1.76635	635.153
11,404	1,317.3	17.9805	1,058.08	1.7844	638.691
11,529	1,317.53	18.0434	1,068.64	1.67869	642.182
11,654	1,317.73	18.1082	1,079.17	1.63357	645.654
11,780	1,317.94	18.1829	1,089.83	1.66337	649.135
11,905	1,318.16	18.2568	1,100.35	1.6539	652.569
12,030	1,318.38	18.3377	1,110.87	1.68389	655.986
12,156	1,318.59	18.4108	1,121.53	1.69411	659.413
12,281	1,318.78	18.4597	1,132.1	1.69094	662.795
12,406	1,318.98	18.531	1,142.65	1.76593	666.159

End of preview.

YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

Reproduction bundle — The (Marginal) Value of a Search Ad

Independent reproduction of ICML 2026 paper UflglraWRa — The (Marginal) Value of a Search Ad: An Online Causal Framework for Repeated Second-price Auctions (Wen, Hu, Han, Yao, Zhou), arXiv:2605.01756.

No official code was released; this is a clean-room re-implementation of the paper's model (Section 2), the interval-splitting HOB estimator (Section 3), the causal IPW + UCB bidding algorithm (Algorithm 1′, Section 5), and the LinUCB baseline of Figure 2. Pure-CPU simulation — the paper has no neural-network component.

Layout

src/env.py        Repeated second-price auction with linear causal ad value (Sec. 2)
src/algos.py      IntervalSplittingCDF (Sec. 3), CausalUCB (Alg. 1'), LinUCBOverbid, SPA/FPA estimators
src/run_all.py    Runs every claim's experiment -> outputs/ (CSVs, Plotly HTML, summary.json)
outputs/          All generated results, figures and summary.json
paper.pdf         The paper

Rerun

python3 src/run_all.py          # full run (~100 s on a laptop CPU)
python3 src/run_all.py --fast   # quick smoke run

Results (see `outputs/summary.json`)

Claim	What we measured	Result
1 (Thm 1)	causal-UCB regret vs T	exponent 0.57 ≈ √T; flattens; beats LinUCB (linear, exp 0.93)
2 (Thm 2)	regret with G known	still √T (exponent 0.57) — barrier persists, as Thm 2 states
3	SPA vs FPA HOB-CDF error, binary	SPA slope −0.52 (→√T) vs FPA −0.30 (→T^{2/3})
4	SPA vs FPA, full-information	both slope −0.52 — payment-rule advantage vanishes
5 (Sec 3)	interval-splitting recovers G	sup-error 0.022 (uniform), 0.007 (trunc-normal); Fig-1 monotonicity corr −1.0

Reproduction target scale matches the paper's own Figure 2 (T = 50 000, 10 seeds). HF Jobs was unavailable (402 — insufficient credits), so the scaled run was executed locally.