Dataset Viewer
The dataset viewer is not available for this split.
Cannot load the dataset split (in streaming mode) to extract the first rows.
Error code: StreamingRowsError
Exception: CastError
Message: Couldn't cast
checks: struct<IPGS is a probability kernel for all gamma>0, K>=2: struct<pass: bool, detail: string>, IPGS (... 1127 chars omitted)
child 0, IPGS is a probability kernel for all gamma>0, K>=2: struct<pass: bool, detail: string>
child 0, pass: bool
child 1, detail: string
child 1, IPGS satisfies p_t(b_t) >= 1/K (used in Part I of Eq. C.15): struct<pass: bool, detail: string>
child 0, pass: bool
child 1, detail: string
child 2, Organic recommendations do not increment N_i or S_i: struct<pass: bool, detail: string>
child 0, pass: bool
child 1, detail: string
child 3, Both stages of the mechanism are exercised: struct<pass: bool, detail: string>
child 0, pass: bool
child 1, detail: string
child 4, Every arm reaches the saturation threshold N before Stage 2: struct<pass: bool, detail: string>
child 0, pass: bool
child 1, detail: string
child 5, RASC explores with frequency 1/L: struct<pass: bool, detail: string>
child 0, pass: bool
child 1, detail: string
child 6, Exploitation starts at epoch m0 = ceil(2 + log2 N): struct<pass: bool, detail: string>
child 0, pass: bool
child 1, detail: string
child 7, Cold-start DBIC bound Eq. (C.7) holds at the prescribed L: struct<pass: bool, detail: string>
child 0, pass: bool
child 1, detail: string
child 8, Cold start is linear in K (paper: O(K L N)): struct<pass: bool, detail: string>
child 0, pass: bool
child 1, de
...
child 4, slope_total: double
child 5, slope_exploit: double
child 6, slope_Tcold: double
wall_clock_s: double
bound_check: struct<formula: string, sigma: double, delta: double, m0: int64, tau_m0_minus_1: double, n_points: i (... 203 chars omitted)
child 0, formula: string
child 1, sigma: double
child 2, delta: double
child 3, m0: int64
child 4, tau_m0_minus_1: double
child 5, n_points: int64
child 6, n_violations: int64
child 7, min_slack: double
child 8, max_slack: double
child 9, median_slack: double
child 10, rows: list<item: struct<T: int64, K: int64, d: int64, seed: int64, measured: double, bound: double, slack: (... 9 chars omitted)
child 0, item: struct<T: int64, K: int64, d: int64, seed: int64, measured: double, bound: double, slack: double>
child 0, T: int64
child 1, K: int64
child 2, d: int64
child 3, seed: int64
child 4, measured: double
child 5, bound: double
child 6, slack: double
rows: list<item: struct<T: int64, K: int64, d: int64, seed: int64, total: double, exploit: double, Tcold: (... 20 chars omitted)
child 0, item: struct<T: int64, K: int64, d: int64, seed: int64, total: double, exploit: double, Tcold: int64, tag: (... 8 chars omitted)
child 0, T: int64
child 1, K: int64
child 2, d: int64
child 3, seed: int64
child 4, total: double
child 5, exploit: double
child 6, Tcold: int64
child 7, tag: string
to
{'meta': {'seeds': Value('int64'), 'N': Value('int64'), 'Tmax': Value('int64'), 'setting': Value('string')}, 'fits': {'T:K3_d5': {'x': List(Value('int64')), 'total': List(Value('float64')), 'exploit': List(Value('float64')), 'Tcold': List(Value('float64')), 'slope_total': Value('float64'), 'slope_exploit': Value('float64'), 'slope_Tcold': Value('float64')}, 'T:K5_d5': {'x': List(Value('int64')), 'total': List(Value('float64')), 'exploit': List(Value('float64')), 'Tcold': List(Value('float64')), 'slope_total': Value('float64'), 'slope_exploit': Value('float64'), 'slope_Tcold': Value('float64')}, 'T:K10_d5': {'x': List(Value('int64')), 'total': List(Value('float64')), 'exploit': List(Value('float64')), 'Tcold': List(Value('float64')), 'slope_total': Value('float64'), 'slope_exploit': Value('float64'), 'slope_Tcold': Value('float64')}, 'T:K5_d10': {'x': List(Value('int64')), 'total': List(Value('float64')), 'exploit': List(Value('float64')), 'Tcold': List(Value('float64')), 'slope_total': Value('float64'), 'slope_exploit': Value('float64'), 'slope_Tcold': Value('float64')}, 'K': {'x': List(Value('int64')), 'total': List(Value('float64')), 'exploit': List(Value('float64')), 'Tcold': List(Value('float64')), 'slope_total': Value('float64'), 'slope_exploit': Value('float64'), 'slope_Tcold': Value('float64'), 'env_gap': List(Value('float64')), 'slope_env_gap': Value('float64'), 'exploit_normalised': List(Value('float64')), 'slope_exploit_normalised': Value('float64')}, 'd': {'x': List(Value('int64')), 'total': List(Value('float64')), 'exploit': List(Value('float64')), 'Tcold': List(Value('float64')), 'slope_total': Value('float64'), 'slope_exploit': Value('float64'), 'slope_Tcold': Value('float64')}}, 'bound_check': {'formula': Value('string'), 'sigma': Value('float64'), 'delta': Value('float64'), 'm0': Value('int64'), 'tau_m0_minus_1': Value('float64'), 'n_points': Value('int64'), 'n_violations': Value('int64'), 'min_slack': Value('float64'), 'max_slack': Value('float64'), 'median_slack': Value('float64'), 'rows': List({'T': Value('int64'), 'K': Value('int64'), 'd': Value('int64'), 'seed': Value('int64'), 'measured': Value('float64'), 'bound': Value('float64'), 'slack': Value('float64')})}, 'rows': List({'T': Value('int64'), 'K': Value('int64'), 'd': Value('int64'), 'seed': Value('int64'), 'total': Value('float64'), 'exploit': Value('float64'), 'Tcold': Value('int64'), 'tag': Value('string')}), 'wall_clock_s': Value('float64')}
because column names don't match
Traceback: Traceback (most recent call last):
File "/src/services/worker/src/worker/utils.py", line 149, in get_rows_or_raise
return get_rows(
dataset=dataset,
...<4 lines>...
column_names=column_names,
)
File "/src/libs/libcommon/src/libcommon/utils.py", line 272, in decorator
return func(*args, **kwargs)
File "/src/services/worker/src/worker/utils.py", line 129, in get_rows
rows_plus_one = list(itertools.islice(safe_iter(ds, dataset=dataset), rows_max_number + 1))
File "/src/services/worker/src/worker/utils.py", line 489, in safe_iter
yield from ds.decode(False) if ds.features else ds
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2818, in __iter__
for key, example in ex_iterable:
^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2355, in __iter__
for key, pa_table in self._iter_arrow():
~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2380, in _iter_arrow
for key, pa_table in self.ex_iterable._iter_arrow():
~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 536, in _iter_arrow
for key, pa_table in iterator:
^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 419, in _iter_arrow
for key, pa_table in self.generate_tables_fn(**gen_kwags):
~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 343, in _generate_tables
self._cast_table(pa_table, json_field_paths=json_field_paths),
~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 132, in _cast_table
pa_table = table_cast(pa_table, self.info.features.arrow_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
checks: struct<IPGS is a probability kernel for all gamma>0, K>=2: struct<pass: bool, detail: string>, IPGS (... 1127 chars omitted)
child 0, IPGS is a probability kernel for all gamma>0, K>=2: struct<pass: bool, detail: string>
child 0, pass: bool
child 1, detail: string
child 1, IPGS satisfies p_t(b_t) >= 1/K (used in Part I of Eq. C.15): struct<pass: bool, detail: string>
child 0, pass: bool
child 1, detail: string
child 2, Organic recommendations do not increment N_i or S_i: struct<pass: bool, detail: string>
child 0, pass: bool
child 1, detail: string
child 3, Both stages of the mechanism are exercised: struct<pass: bool, detail: string>
child 0, pass: bool
child 1, detail: string
child 4, Every arm reaches the saturation threshold N before Stage 2: struct<pass: bool, detail: string>
child 0, pass: bool
child 1, detail: string
child 5, RASC explores with frequency 1/L: struct<pass: bool, detail: string>
child 0, pass: bool
child 1, detail: string
child 6, Exploitation starts at epoch m0 = ceil(2 + log2 N): struct<pass: bool, detail: string>
child 0, pass: bool
child 1, detail: string
child 7, Cold-start DBIC bound Eq. (C.7) holds at the prescribed L: struct<pass: bool, detail: string>
child 0, pass: bool
child 1, detail: string
child 8, Cold start is linear in K (paper: O(K L N)): struct<pass: bool, detail: string>
child 0, pass: bool
child 1, de
...
child 4, slope_total: double
child 5, slope_exploit: double
child 6, slope_Tcold: double
wall_clock_s: double
bound_check: struct<formula: string, sigma: double, delta: double, m0: int64, tau_m0_minus_1: double, n_points: i (... 203 chars omitted)
child 0, formula: string
child 1, sigma: double
child 2, delta: double
child 3, m0: int64
child 4, tau_m0_minus_1: double
child 5, n_points: int64
child 6, n_violations: int64
child 7, min_slack: double
child 8, max_slack: double
child 9, median_slack: double
child 10, rows: list<item: struct<T: int64, K: int64, d: int64, seed: int64, measured: double, bound: double, slack: (... 9 chars omitted)
child 0, item: struct<T: int64, K: int64, d: int64, seed: int64, measured: double, bound: double, slack: double>
child 0, T: int64
child 1, K: int64
child 2, d: int64
child 3, seed: int64
child 4, measured: double
child 5, bound: double
child 6, slack: double
rows: list<item: struct<T: int64, K: int64, d: int64, seed: int64, total: double, exploit: double, Tcold: (... 20 chars omitted)
child 0, item: struct<T: int64, K: int64, d: int64, seed: int64, total: double, exploit: double, Tcold: int64, tag: (... 8 chars omitted)
child 0, T: int64
child 1, K: int64
child 2, d: int64
child 3, seed: int64
child 4, total: double
child 5, exploit: double
child 6, Tcold: int64
child 7, tag: string
to
{'meta': {'seeds': Value('int64'), 'N': Value('int64'), 'Tmax': Value('int64'), 'setting': Value('string')}, 'fits': {'T:K3_d5': {'x': List(Value('int64')), 'total': List(Value('float64')), 'exploit': List(Value('float64')), 'Tcold': List(Value('float64')), 'slope_total': Value('float64'), 'slope_exploit': Value('float64'), 'slope_Tcold': Value('float64')}, 'T:K5_d5': {'x': List(Value('int64')), 'total': List(Value('float64')), 'exploit': List(Value('float64')), 'Tcold': List(Value('float64')), 'slope_total': Value('float64'), 'slope_exploit': Value('float64'), 'slope_Tcold': Value('float64')}, 'T:K10_d5': {'x': List(Value('int64')), 'total': List(Value('float64')), 'exploit': List(Value('float64')), 'Tcold': List(Value('float64')), 'slope_total': Value('float64'), 'slope_exploit': Value('float64'), 'slope_Tcold': Value('float64')}, 'T:K5_d10': {'x': List(Value('int64')), 'total': List(Value('float64')), 'exploit': List(Value('float64')), 'Tcold': List(Value('float64')), 'slope_total': Value('float64'), 'slope_exploit': Value('float64'), 'slope_Tcold': Value('float64')}, 'K': {'x': List(Value('int64')), 'total': List(Value('float64')), 'exploit': List(Value('float64')), 'Tcold': List(Value('float64')), 'slope_total': Value('float64'), 'slope_exploit': Value('float64'), 'slope_Tcold': Value('float64'), 'env_gap': List(Value('float64')), 'slope_env_gap': Value('float64'), 'exploit_normalised': List(Value('float64')), 'slope_exploit_normalised': Value('float64')}, 'd': {'x': List(Value('int64')), 'total': List(Value('float64')), 'exploit': List(Value('float64')), 'Tcold': List(Value('float64')), 'slope_total': Value('float64'), 'slope_exploit': Value('float64'), 'slope_Tcold': Value('float64')}}, 'bound_check': {'formula': Value('string'), 'sigma': Value('float64'), 'delta': Value('float64'), 'm0': Value('int64'), 'tau_m0_minus_1': Value('float64'), 'n_points': Value('int64'), 'n_violations': Value('int64'), 'min_slack': Value('float64'), 'max_slack': Value('float64'), 'median_slack': Value('float64'), 'rows': List({'T': Value('int64'), 'K': Value('int64'), 'd': Value('int64'), 'seed': Value('int64'), 'measured': Value('float64'), 'bound': Value('float64'), 'slack': Value('float64')})}, 'rows': List({'T': Value('int64'), 'K': Value('int64'), 'd': Value('int64'), 'seed': Value('int64'), 'total': Value('float64'), 'exploit': Value('float64'), 'Tcold': Value('int64'), 'tag': Value('string')}), 'wall_clock_s': Value('float64')}
because column names don't matchNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
Reproduction bundle — Incentivized Exploration with Stochastic Covariates: A Two-Stage Mechanism Design for Recommender System
ICML 2026, OpenReview LTHHiPNbrs, arXiv 2406.04374.
Independent from-scratch reproduction: no official code, data pipeline, or checkpoints were released by the authors.
Layout
rcb/
core.py RCB: Algorithm 1 (Cold Start: MPASC -> RASC), Algorithm 2
(Exploitation: inverse proportional gap sampling), the Theorem 1
calibration constants N(eps), L(eps), m0(eps), the Corollary 1
ridge generalization error E_F, and the Gaussian arm posteriors.
env.py SyntheticEnv (Appendix F.1 Settings 1-4, stochastic covariates on
the unit sphere) and build_warfarin (IWPC/PharmGKB feature pipeline
per Appendix F.4: 5528 patients x 70 features).
scripts/
theory_checks.py Executable re-derivation of the Theorem 2 lemma chain
(Lemma 7, Corollary 1, Lemma 11, Eq. E.29/E.30) and of
Theorem 1's four advertised exponents. Exits non-zero on
any failed assertion.
claim3_structure.py Structural invariants of Algorithms 1-2 plus measured
complexity exponents (cold-start length vs K/L/N, per-round
cost vs d, oracle solve vs d).
run_synthetic.py Appendix F.1 Settings 1-4 at the paper's stated scale.
regret_rate_fit.py Claim 1: log-log exponent fits of TOTAL vs EXPLOITATION-stage
regret in T, K, d (Theorem 2 bounds only the latter by
sqrt(Kd(T-T_cold))).
run_warfarin.py Section 5.1 full grid: 3 budgets x 3 prior variances x 10
patient-arrival permutations, T = 5528 each.
warfarin_ablation.py Sensitivity over the four under-specified choices in
Section 5.1 (reward definition, E_F source, phi0, cold-start N)
plus the offline full-information oracle ceiling.
warfarin_gamma_sweep.py Exploration-aggressiveness sweep on the spread parameter.
figures.py Plotly figures; every figure writes its raw data as CSV.
data/ IWPC_Data_8-9-09.xls (public PharmGKB release) and the derived CSVs.
outputs/ JSON results + stdout logs for every run in the logbook.
figs/ figure HTML + the CSV behind each figure.
Rerun
pip install numpy pandas scikit-learn plotly xlrd
cd repro_rcb
python3 scripts/theory_checks.py # ~25 s, exits 0 iff every lemma check passes
python3 scripts/claim3_structure.py # ~2 min
python3 scripts/regret_rate_fit.py --seeds 5 --workers 10
python3 scripts/run_warfarin.py --C_N 1.0 --N 5 --reward binary --EF theory \
--phi0 1.0 --perms 10 --out outputs/warfarin.json
python3 scripts/warfarin_ablation.py --perms 5
python3 scripts/run_synthetic.py --only s1 # Appendix F Setting 1 (slowest, ~1 h)
python3 scripts/figures.py
All results in the logbook were produced on CPU only (16-core x86-64, WSL2 Linux 6.18). No GPU is required and none was used.
Data provenance
data/IWPC_Data_8-9-09.xls is the public International Warfarin Pharmacogenetics
Consortium release distributed by PharmGKB
(https://www.pharmgkb.org/downloads), the same source cited by the paper
(Consortium, 2009). rcb/env.py:build_warfarin reproduces Appendix F.4's feature
construction and yields 5528 patients x 70 features, matching the paper's stated
dimensions exactly.
- Downloads last month
- 54