Datasets:

punctualprocrastinator
/

tabiclv2-prior-exp2

Error code:   StreamingRowsError
Exception:    ValueError
Message:      Dataset 'max_num_classes' has length 1 but expected 14573
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 544, 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 3052, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2559, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2584, 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 558, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 440, 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/hdf5/hdf5.py", line 76, in _generate_tables
                  num_rows = _check_dataset_lengths(h5, self.info.features)
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/hdf5/hdf5.py", line 353, in _check_dataset_lengths
                  raise ValueError(f"Dataset '{path}' has length {dset.shape[0]} but expected {num_rows}")
              ValueError: Dataset 'max_num_classes' has length 1 but expected 14573

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.

TabICLv2 prior tables for a controlled component study

This dataset contains synthetic classification tables only. There is no real-world data in it. The tables were drawn from TabICLv2's own prior, which is the graph_scm prior in soda-inria/tabicl at commit c91f00d, using the prior flags of the official classification recipes scripts/train_v2_clf_stage{1,2}.sh.

Every model in our Stage 1 rerun trains on exactly these files, so all variants see identical tables in identical order. The study retrains small TabICLv2-style models and switches one architectural component at a time.

Files

File	Tables	Content	sha256
`v2_s1.h5`	145,464	TabICLv2 stage 1: 1,024 rows, 1–100 features, 30–90% of rows for training, groups of 4 tables sharing length and split	`52df2293c353c0efb1d4ae2f28c019bf96e1fb4cb3d3d391926f412d1dd940fa`
`v2_long.h5`	14,573	Long stage: TabICLv2's stage-2 flags with the longest table capped at 4,096 rows (400–4,096 rows, log-uniform, 79–81% for training, single-table groups)	`e9701ac013ff091a61f572fee5bbc8d0506df7cb3f2a0d7f8e164e79f02bf9a5`

Caps: every table has at most 10 classes and 100 feature columns. Narrower groups are zero-padded.
Seeds: stage 1 comes from 19 shards with seeds 7000–7018 (1,914 groups each). The long stage comes from 19 shards with seeds 7300–7318 (767 tables each).
Why the long stage is capped: the generator is not bit-identical across machines, so these files are the single shared copy. The paper's 10,240- and 60,000-row stages were left out because they did not fit our GPU budget.

Format

The files are HDF5 dumps in TFM-Playground's prior-dump layout:

X has shape (tables, rows, 100) and y has shape (tables, rows); both are zero-padded.
num_features, num_datapoints and train_test_split_index are stored per table.
original_batch_size is the group size.
problem_type is classification.

TFM-Playground's DumpPrior reads the files directly.

Provenance

The tables were generated with tabicl (BSD-3-Clause) through examples/exp2_dump_tabiclv2.py in our TFM-Playground fork, on an RTX PRO 6000 sandbox, on 2026-10-27.

Downloads last month: -

Total file size:

22.5 GB