The dataset viewer is not available for this split.
Error code: StreamingRowsError
Exception: ValueError
Message: Expected object or value
Traceback: Traceback (most recent call last):
File "/src/services/worker/src/worker/utils.py", line 147, 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 127, 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 483, 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 2951, in __iter__
for key, example in ex_iterable:
^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 517, in __iter__
yield from self.ex_iterable
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 409, in __iter__
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 281, in _generate_tables
examples = [ujson_loads(line) for line in batch.splitlines()]
~~~~~~~~~~~^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/utils/json.py", line 20, in ujson_loads
return pd.io.json.ujson_loads(*args, **kwargs)
~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
ValueError: Expected object or valueNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
Sismo benchmark: run outputs
The outputs of every series of the Sismo public benchmark, by Limiar Research. The code, the protocol, the approval record and the table are in the benchmark repository: https://github.com/LimiarResearch/sismo.
Sismo was compared with 15 change and anomaly detection methods on 2,047 series: synthetic streams, the public NAB, SMD, SMAP and MSL datasets, and a simulated company. A development run came first (protocol v1); the approval protocol (v2) was frozen after it and before the approval run, and the approval seeds were drawn from the hash of the frozen approval files. The approval run stopped twice and was resumed, and it was run again in full with the publication code, with identical results. The verdict of the protocol is "not approved": 10 of the 18 rows that decide passed. Every row is published, the failed ones included.
Seven independent reviews examined this material after the approval run. Their reports and the list of every point they raised are in
paineis/ of the repository, and the points that concern how to read the table are in its README ("Known limitations and caveats").
Errors found after publication are in ERRATA.md of the repository.
Files
| File | What it is |
|---|---|
approval_original.zip |
The approval run (2026-10-12 and 13), on the frozen code of that date, with two stops and resumptions (frozen file sha256 ed41901c...). The code and its output schema were in Portuguese at the time: folder names, JSON keys, method names and row labels are the original ones; protocol/publication/keymap.json in the repository maps every old name to the new one. |
publication.zip |
The same approval run again (2026-10-14), with the English publication code (frozen file sha256 7de3ed49...) and the same seeds. All 2,047 series are identical to the approval run (comparison.txt). |
lives.zip |
The six simulated company lives both runs used (seeds 746892, 155069, 479019; the three null lives with the first). Both runs generated them from the same seeds and they are identical, except the machine timings in custo.json; this copy is the one of the publication run. |
watch_sensitivity.zip |
Information only, outside the verdict (ERRATA.md, entry 1, of the repository): the 1,000 approval synthetic outputs of publication.zip with two more configurations of WATCH and CTM (the martingale threshold alone, and the strictest point of the grid). sha256 a6278032eab29adc5ba64e412ba78455914bd26b89df716562794c19a720067f. No row of the table changes. |
Inside each run archive:
- one gzipped JSON per series, by base (
approval_original/: sinteticos, nab, smd, smap, msl, empresa;publication/: synthetic, nab, smd, smap, msl, company), with the warning lists of every method and configuration and the records and events of the detector; - the rows (
linhas.json/rows.json), the NAB score (nab.json) and the table (tabela.mdandtable.mdin the approval run,table.txtin the publication run); - the run log (
registro.txt/log.txt) and the list of finished series (retomada.txt/resume.txt); - in
publication/,comparison.txt: the full output of the two comparison tools.
Verify
From the root of the repository, with the three archives unpacked next to it:
python protocol/publication/compare_runs.py approval_original publication
python protocol/publication/compare_rows.py approval_original/linhas.json publication/rows.json --nab approval_original/nab.json publication/nab.json
The first compares every series of the two runs; the second compares the rows and the NAB score. To recompute the rows of the approval run in the current schema without running any method:
python protocol/publication/compare_runs.py --convert approval_original approval_converted
python -m metrics.rows approval_converted --lives lives --json rows.json
Notes
approval_original/registro.txtis the run log of the approval, in Portuguese. It records the two stops of the run (the machine was restarted at 20:27 and 22:40 UTC) and the resumptions with the same command. Two annotation lines were shortened for publication: "o contêiner" became "a máquina", and a remark on the probable cause was removed. No timestamp or count was changed, and the log is not part of any hash.- The simulated company's process names, field names and ground truth are in Portuguese; they are data. The glossary is in
generators/company/README.mdof the repository. - The public data (NAB, SMD, SMAP, MSL) are not included. The outputs name the series and hold the warnings of each method, not the data values; the repository's reproduction command downloads the data, each under its own license.
- Downloads last month
- 46