Pith. sign in

REVIEW 3 major objections 7 minor 58 references

Automated, Unsupervised, and Auto-parameterized Inference of Data Patterns and Anomaly Detection

T0 review · 3 major / 7 minor · reviewed 2026-08-11 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read RIOLU claims that a fully automated, unsupervised pipeline can infer precise per-column regular-expression patterns and detect anomalies, with no labeled samples or parameter tuning.

desk verdict Solid unsupervised regex-inference system with honest evaluation, but headline claims need tightening and the 'healthy patterns are frequent' assumption needs a prominent caveat. read the letter →

arxiv 2412.05240 v1 pith:O6OCHJ5J submitted 2024-12-06 cs.SE cs.DB

Qiaolin Qin , Heng Li , Ettore Merlo , Maxime Lamothe This is my paper · ORCID
classification cs.SEcs.DB
keywords datapatterninferenceregexsynthesisanomalydetectionunsupervisedlearningprofilingqualitycoveragerateestimationK-Meansselection
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

RIOLU is a method for learning the syntactic patterns of a data column directly from unlabeled, often dirty input. Its central claim is that the parameters other tools ask users to set—the percentage of healthy values, the granularity of template splitting, and which candidate patterns to keep—can all be estimated automatically from the data itself. If true, data profiling and pattern-anomaly detection would no longer require per-domain regular expressions, frequency thresholds, or labeled examples. The paper reports 97.2% average F1 across 63 profiling datasets, exceeding FlashProfile (96.2%) and ChatGPT (86.4%), and reports large F1 gains over FlashProfile for anomaly detection on five public datasets, with a user-guided variant adding up to 37.4% more F1. The practical stake is that data preparation and quality assurance, often a highly manual stage of software and data engineering, could run with zero configuration on a new column.

What carries the argument

The load-bearing mechanism is the automatic coverage-rate estimator feeding a four-layer constraint waterfall. First, a z-score-based sample size is drawn; then the estimated health fraction $r_{cov}$ controls both how many delimiters are used to split records into raw templates (exact matching rate $r_{EM}=r_{cov}$, following the minimum description length principle) and which token contents and character slots are common enough to become constraints. Constraints are inferred in strict-to-loose order—token range, token length, static character, static character type—so a stricter constraint, once found, prevents looser ones from over-generalizing. Finally, patterns compete by matching rate on the full column, and a two-cluster K-Means split labels the high-frequency cluster as the healthy pattern set, which simultaneously selects patterns and estimates how much of the column is covered. The same mechanism yields patterns for profiling when $r_{cov}=1$ and skips selection.

What would settle it

Take a column where the legitimate format is rare while one malformed format dominates, such as 2% well-formed IPv6 addresses among many copies of a single invalid string. RIOLU's K-Means pattern selection would place the true format in the low-frequency cluster and declare the healthy rows anomalous; comparing its predicted anomalies against hand-labeled ground truth on such a column would settle whether the frequency-based coverage and selection assumptions hold.

Watch

Extended reading notes

Core claim

The paper's discovery is that a rule-based four-layer constraint inference can be made fully unsupervised and auto-parameterized by tying every threshold to one estimated quantity, the coverage rate $r_{cov}$, the fraction of healthy values in a column. RIOLU samples a statistically representative subset, generates raw templates whose splitting granularity is controlled by $r_{cov}$, infers constraints in waterfall order (token range, token length, static character, character type), and selects healthy patterns by K-Means clustering on pattern frequency over the whole column. In the fully automated variant, $r_{cov}$ itself is estimated by generating five initial pattern pools at an assumed 95% coverage, keeping the high-frequency patterns from each pool, and averaging their matching rates on the full column. The evaluation claims 97.2% average F1 on profiling, anomaly-detection F1 between 0.611 and 1.0 across five public datasets, lower average inference time than ChatGPT, and successful use on industrial tables; the paper also states that legitimate patterns occurring in a statistical minority can be mis-flagged as anomalies.

Load-bearing premise

The pipeline assumes that in any column the healthy records are a large, cohesive majority, so legitimate patterns always form high-frequency clusters and anomalies always form low-frequency ones; if valid values are rare, scattered, or heterogeneous, the estimated coverage rate shrinks and healthy records get flagged as anomalies.

Editorial extensions

If this is right

  • A new data column can be profiled and checked for pattern violations with zero configuration, removing per-dataset thresholds such as the low-frequency cutoff that FlashProfile exposes.
  • The same run that produces the patterns also produces an automatic estimate of the column's error rate, so data teams get a health report rather than only a list of anomalies.
  • A small labeled subset can be traded for higher precision: Guided-RIOLU reports up to 37.4% F1 improvement over the automated version using under 0.4% of rows as labels.
  • Because the fully automated inference is fast in the paper's measurements, pattern validation could be embedded in continuous data pipelines rather than run as a one-off preparation step.
  • Columns with heterogeneous legitimate formats, such as email fields with and without a dot separator, remain the weak spot the paper identifies for the automated version.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A natural test the paper does not run is to calibrate the unsupervised $r_{cov}$ estimator on columns with known error rates spanning the 0.2% to 52.5% range; the Flights results suggest the estimator degrades as healthy patterns fragment, so a bias curve would show where the automated mode can be trusted.
  • The same frequency-split logic could be extended to heterogeneous columns by clustering templates on syntactic or semantic similarity before the frequency split, which would address the email-field case the paper flags as hard.
  • If the pattern pool includes a very broad fallback pattern, the K-Means selection could mistake over-generalization for health; adding a precision check against out-of-domain strings, as in the profiling evaluation, would be a direct extension.
  • The auto-parameterization claim is scoped to regex syntax; extending the approach to typed values (treating dates as dates, not strings) would require semantic constraints outside the four-layer waterfall.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 7 minor

Summary. The paper introduces RIOLU, an unsupervised and supposedly auto-parameterized approach for inferring regular-expression patterns from data columns and for detecting pattern anomalies. The pipeline consists of column sampling, coverage-rate estimation (supervised or unsupervised), constrained template generation, four-layer constraint inference (token range, token length, static character, static type), and K-Means-based pattern selection. The data-profiling variant is evaluated on 63 FlashProfile-DOMAINS datasets and compared with FlashProfile and ChatGPT; the anomaly-detection variant is evaluated on five public datasets with ground-truth error labels, on method-name consistency in 11 Java projects, and in an industrial deployment at CompanyX. The paper reports that Auto-RIOLU achieves 97.2% average F1 for profiling, outperforms the baselines on four of five anomaly-detection datasets, and that Guided-RIOLU further improves F1 with a small labeled subset.

Significance. If the results hold, RIOLU would be a practically useful contribution: it targets a real pain point (manual configuration and labeled data for pattern inference), and its design choices are evaluated with held-out training/test splits, cross-domain false-positive measurement, an ablation study, a sensitivity analysis, five-run averages for the anomaly-detection experiments, and a public replication package. The paper is also candid about some failure modes, notably scattered healthy patterns in the Flights dataset. However, the central claim of fully automatic, domain-general anomaly detection rests on a frequency-based assumption about healthy versus anomalous patterns, and that assumption is both explicitly conceded as limited in Section VI and empirically violated in one of the paper's own datasets. The abstract's unqualified 'various domains' claim is therefore stronger than the evidence supports, and the per-dataset comparison with ChatGPT is weaker than the average-based headline suggests.

major comments (3)
  1. [Section III-E, III-F, IV-B.4, VI] The unsupervised coverage-rate estimator and the pattern-selection step both assume that healthy records form high-frequency pattern clusters while anomalies form low-frequency clusters. This assumption is load-bearing for the central claim of fully automatic anomaly detection, and the paper itself concedes in Section VI that 'records with rare legitimate patterns may be flagged due to their statistical minority.' The Flights dataset is a concrete instance: Section IV-B.4 reports that its ground-truth healthy patterns are scattered, and Table VII shows that removing pattern selection collapses Auto-RIOLU's F1 from 0.738 to 0.135. Because the abstract claims unqualified success on 'datasets in various domains,' the paper should either qualify that claim to exclude or explicitly bound the scattered-healthy-pattern regime, or provide a mechanism and a decisive test for detecting when the frequency assumption fails.
  2. [Abstract and Table III] The abstract states that RIOLU 'even outperforms ChatGPT in terms of both accuracy (12.3% higher F1) and efficiency,' but this is only an average over five datasets. Per-dataset results in Table III show that Auto-RIOLU does not outperform ChatGPT on Hosp-100k (0.485 vs. 0.550) and is only marginally better on Flights (0.738 vs. 0.733). The paper should report per-dataset differences, clarify that the 12.3% figure is an average, and state which baseline the 'up to 800.4% improvement' refers to. This is not just a presentation issue, because the abstract's phrasing can mislead readers about the consistency of the improvement.
  3. [Section IV-B.3 and Table III] The anomaly-detection results in Table III are reported as averages over five runs, but no standard deviations, confidence intervals, or per-run values are given, despite Section VI acknowledging that random sampling may cause fluctuation. Given that several comparisons are close (e.g., Auto-RIOLU vs. ChatGPT on Flights), the absence of variance information makes it impossible to assess whether the reported differences are stable. I request that the authors add dispersion measures (e.g., standard deviation or min-max range) for the five runs, at least for the F1 scores that support the headline claims.
minor comments (7)
  1. [Section II-C] There is a typo: 'abnomal' should be 'anomaly'.
  2. [Section III-F] The phrase 'initial overage rate' should be 'initial coverage rate' (and the same typo appears in the caption of Figure 4).
  3. [Section IV-A.4] The text says 'the patterns created using our raw templates and water flow constraint selection approaches'; this should presumably be 'waterfall constraint selection' to match the description in Section III-D.
  4. [Related Work] The phrase 'pattern-bassed anomaly detection' contains a typo; it should be 'pattern-based anomaly detection'.
  5. [Section IV-B.4, Java evaluation] The sentence 'For one project (liferay-portal), the false positive rate in the detected anomalies is 50%, which may be caused by its low anomaly rate: there are only 4 detected anomalies' appears to refer to the wrong project: in Table VI, liferay-portal has 379 detected anomalies with an FP rate of 0.292, whereas presto has 4 detected anomalies with an FP rate of 0.5. Please correct the project name.
  6. [Section III-B and III-F] The paper claims that RIOLU is 'auto-parameterized,' but several parameters remain fixed by default (sampling confidence level and margin, r_cov_init=0.95, Nsubset=5, and the inserted K-Means anchor frequencies). The sensitivity analysis supports the default choices, but the text should state more precisely that the method is free of user-tunable parameters rather than entirely free of parameters.
  7. [Abstract] The abstract's 'up to 800.4% improvement in terms of F1' is not attributed to a specific dataset or baseline; please add a pointer to the corresponding table and row so the claim can be verified.

Circularity Check

1 steps flagged · score 2.0 of 10

No load-bearing circularity; the unsupervised error-rate estimate is self-referential in definition but the main F1 claims are validated on external ground-truth labels and can fail.

  1. self definitional [Sec. III-E (Pattern Selection) and Sec. III-F (Coverage Rate (rcov) Estimation)]
    "Patterns labeled as high-frequency are selected as healthy patterns, whereas those labeled as low-frequency are not further used. Finally, the healthy patterns are used to detect anomalous records in the column: records that do not match any healthy pattern are identified as anomalies. ... each selected pattern pool is used to calculate the portion of matches they can create (i.e., matching rate) on the whole dataset. ... Hence, we take the average matching rate as the estimated rcov."

    In the unsupervised path, 'healthy' and 'anomaly' are defined by the same K-Means frequency split: high-frequency patterns are healthy, low-frequency patterns are dropped, and any record not matching a healthy pattern is an anomaly. The unsupervised rcov estimator then takes the average matching rate of those selected high-frequency patterns on the same column, so the reported 'error rate' is definitionally the complement of high-frequency coverage; it would equal that even without any external ground truth. The frequency assumption is built into the definition rather than derived from data.

full rationale

The derivation chain is otherwise self-contained. RQ1 uses held-out evaluation: patterns are generated from 20% of each dataset and tested on the remaining 80% plus records from other domains, so the 97.2% F1 claim is not an in-sample fit. RQ2 uses ground-truth labels from prior studies (Raha-era and FlashProfile-DOMAINS datasets), not from RIOLU, and both the ablation study and the Flights failure (F1 dropping from 0.738 to 0.135 without pattern selection) show the method can be wrong, which is the signature of an honest, non-circular test. There is no load-bearing self-citation or imported-uniqueness argument; citations to Auto-Validate, FlashProfile, and Potter's Wheel are external prior work that motivate rather than define RIOLU's result. The one mild circularity is the unsupervised rcov/anomaly operationalization described above: the error-rate estimate and the anomaly definition are two sides of the same high-frequency assumption. Because the evaluation is external and the system can demonstrably fail, this self-reference warrants a score of 2 rather than a higher score.

Assumptions & free parameters 4 free parameters · 6 assumptions · 0 invented entities

RIOLU introduces no new physical or conceptual entity; the inserted K-Means anchor values are algorithmic devices, not entities, so there is no graviton-problem burden. The method's external inputs are hand-set defaults (r_cov_init=0.95, Nsubset=5, the anchors) plus the domain assumption that healthy patterns are always the high-frequency ones. The evaluation additionally inherits the correctness of ground-truth labels from prior data-quality studies. The sensitivity analysis shows the defaults are not sharply tuned, which limits but does not remove the parameter burden.

free parameters (4)
  • r_cov_init = 0.95
    Starting assumption for the unsupervised coverage-rate estimator (Sec III-F), justified by Auto-Validate's claim that anomaly rates are typically below 5% [17]. Varying it from 0.85 to 0.95 changes average F1 by about 3.1%, so it is a stable hand-set default rather than a value fitted to the test data.
  • Nsubset = 5
    Number of sampled subsets used to estimate r_cov (Sec III-F). Chosen because the sensitivity analysis in Sec IV-C shows 'a peak of average F1 score when setting Nsubset to 5' on the five evaluation datasets; this is mild tuning on the test data, though the curve is flat across 1 to 10.
  • Sampling confidence and margin = 95% confidence, 5% margin
    The z-score based sample size Ntr (Sec III-B). A standard statistical convention rather than a fitted value, but it fixes how much data the pattern generator sees and therefore affects which patterns are discoverable.
  • K-Means anchor frequencies = 1 and 1/Ntr inserted into frequency lists; 1/N in pattern selection
    Artificially inserted points to guarantee two clusters exist in the token-range, static-character and pattern-selection tests (Sec III-D.1, III-E). Their handling in the token-range test is not specified, which makes the Fig 3 example outcome not derivable from the text alone.
assumptions (6)
  • domain assumption Healthy records form large high-frequency pattern clusters; anomalies form small low-frequency ones.
    Invoked in pattern selection (Sec III-E: patterns with low matching rates are anomalies) and in the unsupervised r_cov estimator (Sec III-F: 'anomalies cannot form large pattern clusters'). Fails visibly on the Flights dataset and on heterogeneous columns, as the paper concedes in Sec VI.
  • domain assumption Anomaly rates in real-world data are typically below 5%.
    Used to set r_cov_init=0.95 (Sec III-F), citing Auto-Validate [17]. Not verified on the five evaluation datasets, two of which have error rates of 23-53%.
  • domain assumption Two-class K-Means on one-dimensional frequency lists separates healthy from anomalous values without a domain-specific threshold.
    Used in token-range, static-character (Sec III-D.1, III-D.3) and pattern-selection (Sec III-E) decisions; the two artificial anchors guarantee a split exists even when the data do not support one.
  • standard math A random sample of size Ntr at 95% confidence and 5% margin represents the column's pattern distribution.
    Column sampling step (Sec III-B). Standard statistics, but the underlying assumption that the column's patterns are stationary across rows is unstated.
  • standard math Minimum description length guides the choice of how many delimiters to keep.
    Used in constrained template generation (Sec III-C), citing Grunwald [34]. A standard principle, applied here heuristically to pick the maximum number of delimiters under the r_EM constraint.
  • domain assumption Ground-truth labels of the five public datasets from prior studies are correct.
    Sec IV-B.1 obtains 'the original data and their cleaned versions from previous data quality studies'; all anomaly-detection metrics inherit the quality of those external labels.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Automated, Unsupervised, and Auto-parameterized Inference of Data Patterns and Anomaly Detection." pith.science (2026). https://pith.science/paper/O6OCHJ5J

@misc{pith2026241205240,
  author       = {Pith},
  title        = {Pith review of: Automated, Unsupervised, and Auto-parameterized Inference of Data Patterns and Anomaly Detection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/O6OCHJ5J}},
  note         = {Machine review of arXiv:2412.05240}
}
read the original abstract

With the advent of data-centric and machine learning (ML) systems, data quality is playing an increasingly critical role in ensuring the overall quality of software systems. Data preparation, an essential step towards high data quality, is known to be a highly effort-intensive process. Although prior studies have dealt with one of the most impacting issues, data pattern violations, these studies usually require data-specific configurations (i.e., parameterized) or use carefully curated data as learning examples (i.e., supervised), relying on domain knowledge and deep understanding of the data, or demanding significant manual effort. In this paper, we introduce RIOLU: Regex Inferencer auto-parameterized Learning with Uncleaned data. RIOLU is fully automated, automatically parameterized, and does not need labeled samples. RIOLU can generate precise patterns from datasets in various domains, with a high F1 score of 97.2%, exceeding the state-of-the-art baseline. In addition, according to our experiment on five datasets with anomalies, RIOLU can automatically estimate a data column's error rate, draw normal patterns, and predict anomalies from unlabeled data with higher performance (up to 800.4% improvement in terms of F1) than the state-of-the-art baseline, even outperforming ChatGPT in terms of both accuracy (12.3% higher F1) and efficiency (10% less inference time). A variant of RIOLU, with user guidance, can further boost its precision, with up to 37.4% improvement in terms of F1. Our evaluation in an industrial setting further demonstrates the practical benefits of RIOLU.

Figures

Figures reproduced from arXiv: 2412.05240 by the authors.

Figure 1
Figure 1. An overview of RIOLU’s structure. III. APPROACH A. Overview The input of RIOLU is a two-dimensional table with random numbers of columns and rows. The structure of each column is undefined: for example, they can be names, IDs, or URLs. The goal of RIOLU is to automatically derive the patterns (e.g., YYYY-MM-DD) of each column without prior knowledge; the patterns can then be used to detect data anomalies (e.g., inva… view at source ↗
Figure 2
Figure 2. An example of 5 date time records and the templates generated under [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. A date time example of using RIOLU generating patterns ( [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: The coverage rate estimation process in unsupervised approach. [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: RIOLU’s profiling quality on FlashProfile-DOMAINS dataset. The [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: The impact of choosing different rcov init values (0.85, 0.9, 0.95, and 0.99) and Nsubset (range from 1 to 10) for Auto-RIOLU. Evaluation on CompanyX database. The data pattern inconsistency would block the automatic software pipeline and lead to software quality probl…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

58 extracted references · 47 canonical work pages

  1. [1]

    Sommerville, Software Engineering, 10th ed

    I. Sommerville, Software Engineering, 10th ed. Pearson, 2015

    2015

  2. [2]

    Data- centric systems and applications,

    M. Carey, S. Ceri, P. Bernstein, U. Dayal, C. Faloutsos, J. Freytag, G. Gardarin, W. Jonker, V . Krishnamurthy, M. Neimat et al. , “Data- centric systems and applications,” 2008

    work page 2008

  3. [3]

    What is really different in engineering ai-enabled systems?

    I. Ozkaya, “What is really different in engineering ai-enabled systems?” IEEE software, vol. 37, no. 4, pp. 3–6, 2020

    work page 2020

  4. [4]

    A systematic literature review on software defect prediction using artificial intelligence: Datasets, data validation methods, approaches, and tools,

    J. Pachouly, S. Ahirrao, K. Kotecha, G. Selvachandran, and A. Abraham, “A systematic literature review on software defect prediction using artificial intelligence: Datasets, data validation methods, approaches, and tools,” Engineering Applications of Artificial Intelligence , vol. 111, p. 104773, 2022

    work page 2022

  5. [5]

    Data preparation for software vul- nerability prediction: A systematic literature review,

    R. Croft, Y . Xie, and M. A. Babar, “Data preparation for software vul- nerability prediction: A systematic literature review,” IEEE Transactions on Software Engineering , vol. 49, no. 3, pp. 1044–1063, 2022

    2022

  6. [6]

    Are we building on the rock? on the importance of data preprocessing for code summarization,

    L. Shi, F. Mu, X. Chen, S. Wang, J. Wang, Y . Yang, G. Li, X. Xia, and Q. Wang, “Are we building on the rock? on the importance of data preprocessing for code summarization,” in Proceedings of the 30th ACM Joint European Software Engineering Conference and Symposium on the Foundations of Software Engineering , 2022, pp. 107–119

    2022

  7. [7]

    Data quality for software vulnerability datasets,

    R. Croft, M. A. Babar, and M. M. Kholoosi, “Data quality for software vulnerability datasets,” in 2023 IEEE/ACM 45th International Confer- ence on Software Engineering (ICSE) . IEEE, 2023, pp. 121–133

    2023

  8. [8]

    Data quality matters: A case study of obsolete comment detection,

    S. Xu, Y . Yao, F. Xu, T. Gu, J. Xu, and X. Ma, “Data quality matters: A case study of obsolete comment detection,” in 2023 IEEE/ACM 45th International Conference on Software Engineering (ICSE). IEEE, 2023, pp. 781–793

    2023

Show all 58 references
  1. [9]

    Big data systems: A software engineering perspective,

    A. Davoudian and M. Liu, “Big data systems: A software engineering perspective,” ACM Computing Surveys (CSUR), vol. 53, no. 5, pp. 1–39, 2020

    2020

  2. [10]

    Autoconsis: Automatic gui-driven data inconsistency detection of mobile apps,

    Y . Hu, H. Jin, X. Wang, J. Gu, S. Guo, C. Chen, X. Wang, and Y . Zhou, “Autoconsis: Automatic gui-driven data inconsistency detection of mobile apps,” in Proceedings of the 46th International Conference on Software Engineering: Software Engineering in Practice , 2024, pp. 137–146

    2024

  3. [11]

    Dclink: Bridging data constraint changes and implementations in fintech systems,

    W. Tange, C. Wang, P. Yao, R. Wu, X. Fu, G. Fan, and C. Zhang, “Dclink: Bridging data constraint changes and implementations in fintech systems,” in 2023 38th IEEE/ACM International Conference on Automated Software Engineering (ASE) . IEEE, 2023, pp. 914–925

    2023

  4. [12]

    On the expe- riences of adopting automated data validation in an industrial machine learning project,

    L. E. Lwakatare, E. R ˚ange, I. Crnkovic, and J. Bosch, “On the expe- riences of adopting automated data validation in an industrial machine learning project,” in 2021 IEEE/ACM 43rd International Conference on Software Engineering: Software Engineering in Practice (ICSE-SEIP) ...

    2021

  5. [13]

    Machine learning and big data: What is important?

    M. Stonebraker and E. K. Rezig, “Machine learning and big data: What is important?” IEEE Data Eng. Bull. , vol. 42, no. 4, pp. 3–7, 2019

    2019

  6. [14]

    Detecting data errors: Where are we and what needs to be done?

    Z. Abedjan, X. Chu, D. Deng, R. C. Fernandez, I. F. Ilyas, M. Ouz- zani, P. Papotti, M. Stonebraker, and N. Tang, “Detecting data errors: Where are we and what needs to be done?” Proceedings of the VLDB Endowment, vol. 9, no. 12, pp. 993–1004, 2016

    2016

  7. [15]

    Raha: A configuration-free error detec- tion system,

    M. Mahdavi, Z. Abedjan, R. Castro Fernandez, S. Madden, M. Ouzzani, M. Stonebraker, and N. Tang, “Raha: A configuration-free error detec- tion system,” in Proceedings of the 2019 International Conference on Management of Data , 2019, pp. 865–882

    2019

  8. [16]

    Metadata-driven error detection,

    L. Visengeriyeva and Z. Abedjan, “Metadata-driven error detection,” in Proceedings of the 30th International Conference on Scientific and Statistical Database Management , 2018, pp. 1–12

    2018

  9. [17]

    Auto-validate: Unsupervised data validation using data-domain patterns inferred from data lakes,

    J. Song and Y . He, “Auto-validate: Unsupervised data validation using data-domain patterns inferred from data lakes,” in Proceedings of the 2021 International Conference on Management of Data, 2021, pp. 1678– 1691

    2021

  10. [18]

    Log-based anomaly detection with deep learning: How far are we?

    V .-H. Le and H. Zhang, “Log-based anomaly detection with deep learning: How far are we?” in Proceedings of the 44th international conference on software engineering , 2022, pp. 1356–1367

    2022

  11. [19]

    Deeplog: Anomaly detection and diagnosis from system logs through deep learning,

    M. Du, F. Li, G. Zheng, and V . Srikumar, “Deeplog: Anomaly detection and diagnosis from system logs through deep learning,” in Proceedings of the 2017 ACM SIGSAC conference on computer and communications security, 2017, pp. 1285–1298

    2017

  12. [20]

    Pyodds: An end- to-end outlier detection system with automated machine learning,

    Y . Li, D. Zha, P. Venugopal, N. Zou, and X. Hu, “Pyodds: An end- to-end outlier detection system with automated machine learning,” in Companion Proceedings of the Web Conference 2020 , 2020, pp. 153– 157

    2020

  13. [21]

    Tods: An automated time series outlier detection system,

    K.-H. Lai, D. Zha, G. Wang, J. Xu, Y . Zhao, D. Kumar, Y . Chen, P. Zumkhawaka, M. Wan, D. Martinez et al., “Tods: An automated time series outlier detection system,” in Proceedings of the AAAI conference on artificial intelligence , vol. 35, no. 18, 2021, pp. 16 060–16 062

    2021

  14. [22]

    Unsupervised anomaly detection via variational auto- encoder for seasonal KPIs in web applications,

    H. Xu, W. Chen, N. Zhao, Z. Li, J. Bu, Z. Li, Y . Liu, Y . Zhao, D. Pei, Y . Feng et al. , “Unsupervised anomaly detection via variational auto- encoder for seasonal KPIs in web applications,” in Proceedings of the 2018 world wide web conference , 2018, pp. 187–196

    2018

  15. [23]

    Flashprofile: a framework for synthesizing data profiles,

    S. Padhi, P. Jain, D. Perelman, O. Polozov, S. Gulwani, and T. Millstein, “Flashprofile: a framework for synthesizing data profiles,” Proceedings of the ACM on Programming Languages, vol. 2, no. OOPSLA, pp. 1–28, 2018

    2018

  16. [24]

    Data validation for machine learning

    E. Breck, N. Polyzotis, S. Roy, S. Whang, and M. Zinkevich, “Data validation for machine learning.” in MLSys, 2019

    2019

  17. [25]

    Automating large-scale data quality verification,

    S. Schelter, D. Lange, P. Schmidt, M. Celikel, F. Biessmann, and A. Grafberger, “Automating large-scale data quality verification,” Pro- ceedings of the VLDB Endowment, vol. 11, no. 12, pp. 1781–1794, 2018

    2018

  18. [26]

    Potter’s wheel: An interactive data cleaning system,

    V . Raman and J. M. Hellerstein, “Potter’s wheel: An interactive data cleaning system,” in VLDB, vol. 1, 2001, pp. 381–390

    2001

  19. [27]

    Extracting syntactical patterns from databases,

    A. Ilyas, J. M. da Trindade, R. C. Fernandez, and S. Madden, “Extracting syntactical patterns from databases,” in 2018 IEEE 34th International Conference on Data Engineering (ICDE) . IEEE, 2018, pp. 41–52

    2018

  20. [28]

    Human-in- the-loop regular expression extraction for single column format incon- sistency,

    S. Yu, L. Han, M. Indulska, S. Sadiq, and G. Demartini, “Human-in- the-loop regular expression extraction for single column format incon- sistency,” in Proceedings of the ACM Web Conference 2023 , 2023, pp. 3859–3867

    2023

  21. [29]

    The replication package of riolu,

    “The replication package of riolu,” https://github.com/mooselab/ Discover-Data-Quality-With-RIOLU, 2024

    2024

  22. [30]

    Auto-detect: Data-driven error detection in tables,

    Z. Huang and Y . He, “Auto-detect: Data-driven error detection in tables,” in Proceedings of the 2018 International Conference on Management of Data, 2018, pp. 1377–1392

    2018

  23. [31]

    Auto-tag: Tagging-data-by-example in data lakes,

    Y . He, J. Song, Y . Wang, S. Chaudhuri, V . Anil, B. Lassiter, Y . Goland, and G. Malhotra, “Auto-tag: Tagging-data-by-example in data lakes,” arXiv preprint arXiv:2112.06049 , 2021

    2021 arXiv

  24. [32]

    SQL Server Data Tools Documentation,

    Microsoft, “SQL Server Data Tools Documentation,” https://docs. microsoft.com/en-gb/sql/ssdt, [Online; accessed 18-March-2024]

    2024

  25. [33]

    Ataccama Website,

    Ataccama Corporation, “Ataccama Website,” https://www.ataccama. com/, [Online; accessed 18-March-2024]

    2024

  26. [34]

    P. D. Gr ¨unwald, The minimum description length principle . MIT press, 2007

    2007

  27. [35]

    Least squares quantization in pcm,

    S. Lloyd, “Least squares quantization in pcm,” IEEE transactions on information theory, vol. 28, no. 2, pp. 129–137, 1982

    1982

  28. [36]

    Data profiling dataset,

    S. Padhi, “Data profiling dataset,” https://github.com/SaswatPadhi/ FlashProfileDemo/tree/master/tests, 2024, [Online]

    2024

  29. [37]

    Chatgpt,

    OpenAI, “Chatgpt,” https://openai.com/chatgpt/, 2024, [Online]

    2024

  30. [38]

    Understanding regular expression denial of service (redos): Insights from llm-generated regexes and developer forums,

    M. L. Siddiq, J. Zhang, and J. C. D. S. Santos, “Understanding regular expression denial of service (redos): Insights from llm-generated regexes and developer forums,” in Proceedings of the 32nd IEEE/ACM International Conference on Program Comprehension , 2024, pp. 190– 201

    2024

  31. [39]

    Holoclean: Holistic data repairs with probabilistic inference,

    T. Rekatsinas, X. Chu, I. F. Ilyas, and C. R ´e, “Holoclean: Holistic data repairs with probabilistic inference,” arXiv preprint arXiv:1702.00820 , 2017

    2017 arXiv

  32. [40]

    Truth finding on the deep web: Is the problem solved?

    X. Li, X. L. Dong, K. Lyons, W. Meng, and D. Srivastava, “Truth finding on the deep web: Is the problem solved?” arXiv preprint arXiv:1503.00303, 2015

    2015 arXiv

  33. [41]

    Prose pattern inspector document,

    Microsoft, “Prose pattern inspector document,” https://www.microsoft. com/en-us/research/project/prose-pattern-inspector/, 2024, [Online; ac- cessed Jul 16, 2024]

    2024

  34. [42]

    McConnell, Code Complete: Steve McConnell

    S. McConnell, Code Complete: Steve McConnell . Microsoft Press, 1993

    1993

  35. [43]

    Beck, Implementation patterns

    K. Beck, Implementation patterns. Pearson Education, 2007

    2007

  36. [44]

    R. C. Martin, Clean code: a handbook of agile software craftsmanship . Pearson Education, 2009

    2009

  37. [45]

    The effects of comments and identifier names on program comprehensibility: an experimental investigation,

    A. A. Takang, P. A. Grubb, R. D. Macredie et al. , “The effects of comments and identifier names on program comprehensibility: an experimental investigation,” J. Prog. Lang., vol. 4, no. 3, pp. 143–167, 1996

    1996

  38. [46]

    Relating identifier naming flaws and code quality: An empirical study,

    S. Butler, M. Wermelinger, Y . Yu, and H. Sharp, “Relating identifier naming flaws and code quality: An empirical study,” in 2009 16th Working Conference on Reverse Engineering . IEEE, 2009, pp. 31– 35

    2009

  39. [47]

    Java naming conventions,

    Oracle, “Java naming conventions,” https://www.oracle.com/java/ technologies/javase/codeconventions-namingconventions.html, 2024, [Online; accessed Jul 16, 2024]

    2024

  40. [48]

    A convolutional attention network for extreme summarization of source code,

    M. Allamanis, H. Peng, and C. Sutton, “A convolutional attention network for extreme summarization of source code,” in International Conference on Machine Learning (ICML) , 2016

    2016

  41. [49]

    Data smells: categories, causes and consequences, and detection of suspicious data in ai-based systems,

    H. Foidl, M. Felderer, and R. Ramler, “Data smells: categories, causes and consequences, and detection of suspicious data in ai-based systems,” in Proceedings of the 1st International Conference on AI Engineering: Software Engineering for AI , 2022, pp. 229–239

    2022

  42. [50]

    Data smells in public datasets,

    A. Shome, L. Cruz, and A. Van Deursen, “Data smells in public datasets,” in Proceedings of the 1st International Conference on AI Engineering: Software Engineering for AI , 2022, pp. 205–216

    2022

  43. [51]

    Deequ - data quality validation for machine learning pipelines,

    S. Schelter, P. Schmidt, T. Rukat, M. Kiessling, A. Taptunov, F. Biessmann, and D. Lange, “Deequ - data quality validation for machine learning pipelines,” in NeurIPS 2018 ,

    2018

  44. [52]

    Learning regular sets from queries and counterexamples,

    D. Angluin, “Learning regular sets from queries and counterexamples,” Information and computation , vol. 75, no. 2, pp. 87–106, 1987

    1987

  45. [53]

    Synthesizing program input grammars,

    O. Bastani, R. Sharma, A. Aiken, and P. Liang, “Synthesizing program input grammars,” ACM SIGPLAN Notices , vol. 52, no. 6, pp. 95–110, 2017

    2017

  46. [54]

    Mining input grammars from dynamic control flow,

    R. Gopinath, B. Mathis, and A. Zeller, “Mining input grammars from dynamic control flow,” in Proceedings of the 28th acm joint meeting on european software engineering conference and symposium on the foundations of software engineering , 2020, pp. 172–183

    2020

  47. [55]

    Learning highly recursive input grammars,

    N. Kulkarni, C. Lemieux, and K. Sen, “Learning highly recursive input grammars,” in 2021 36th IEEE/ACM International Conference on Automated Software Engineering (ASE) . IEEE, 2021, pp. 456–467

    2021

  48. [56]

    Grammars for free: Toward grammar inference for ad hoc parsers,

    M. Schr ¨oder and J. Cito, “Grammars for free: Toward grammar inference for ad hoc parsers,” in Proceedings of the ACM/IEEE 44th International Conference on Software Engineering: New Ideas and Emerging Results , 2022, pp. 41–45

    2022

  49. [57]

    Deep clustering: A comprehensive survey,

    Y . Ren, J. Pu, Z. Yang, J. Xu, G. Li, X. Pu, S. Y . Philip, and L. He, “Deep clustering: A comprehensive survey,” IEEE Transactions on Neural Networks and Learning Systems , 2024

    2024

  50. [2018]

    Available: https://www.amazon.science/publications/ deequ-data-quality-validation-for-machine-learning-pipelines

    [Online]. Available: https://www.amazon.science/publications/ deequ-data-quality-validation-for-machine-learning-pipelines

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.