Pith. sign in

REVIEW 4 major objections 6 minor 49 references

GPT-FT: An Efficient Automated Feature Transformation Using GPT for Sequence Reconstruction and Performance Enhancement

T0 review · 4 major / 6 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read A decoder-only GPT with a 64-dimensional embedding can be trained to reconstruct feature transformation sequences and estimate their performance, then refined by gradient ascent to generate transformations that beat nine prior automated fea

desk verdict Efficiency gains look real, but the paper's own Table 1 contradicts its central performance claim; needs major revision before it can be trusted. read the letter →

arxiv 2508.20824 v1 pith:2Z4VXX2W submitted 2025-08-28 cs.LG

Yang Gao , Dongjie Wang , Scott Piersall , Ye Zhang , Liqiang Wang This is my paper · ORCID
classification cs.LG
keywords AutomatedfeaturetransformationGPTGradient-ascentsearchEmbeddingspaceSequencereconstructionPerformanceestimationMulti-objectiveoptimizationengineering
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

The paper proposes GPT-FT, a four-stage pipeline that turns automated feature transformation into continuous embedding optimization instead of discrete search. It trains a small decoder-only GPT to both reconstruct transformation sequences and estimate their downstream performance, then runs gradient ascent on the learned embeddings and decodes improved sequences. On 15 benchmark datasets, it reports F1-score and 1-RAE results that match or beat nine earlier methods, including the larger encoder-decoder LSTM model MOAT, with roughly half the parameters and faster inference. The intended payoff is that transformer-based sequence models can make automated feature engineering scalable and cheap enough for practical, even real-time, use.

What carries the argument

The central object is the revised GPT: a single-layer decoder-only transformer with embedding size 64 that maps a postfix transformation sequence to an embedding, reconstructs the sequence through a text predictor head, and estimates its downstream performance through a task classifier head. The joint loss L = αL_pre + (1−α)L_cls trains all heads together, balancing sequence reconstruction against performance estimation. The same embedding space is then searched by gradient ascent (Ẽ = E + η ∂G/∂E) starting from the top-42 seed records, and the optimized embeddings are decoded autoregressively into candidate transformation sequences.

What would settle it

Take a dataset, train GPT-FT with the paper's settings, and compare the true held-out performance of sequences decoded after gradient ascent against the top-42 seed records without gradient ascent. If the best decoded sequence does not beat the best seed more often than random gradient directions do, the central claim fails; the paper's zero-gradient ablation is the natural control.

Watch

Extended reading notes

Core claim

The paper's central claim is that a decoder-only GPT-1-style model with embedding size 64, trained from scratch on reinforcement-learning-collected transformation records, can serve simultaneously as a sequence generator and a performance estimator, and that gradient ascent in this embedding space yields better transformation sequences than the discrete search used by earlier approaches. GPT-FT replaces MOAT's separate encoder, evaluator, and decoder with one compact model trained on a joint loss of reconstruction and performance estimation. The paper asserts that Table 1 shows GPT-FT outperforming all nine baselines across all 15 datasets, with parameter sizes about half of MOAT's and infer

Load-bearing premise

The load-bearing premise is that the embedding space learned from the RL-collected records is smooth and calibrated enough that moving along the gradient of the learned performance head lands on embeddings that decode into valid transformation sequences that genuinely improve real downstream performance, not just the predicted score.

Editorial extensions

If this is right

  • Feature transformation can be formulated as differentiable optimization in a compact learned embedding space, replacing combinatorial search with a few gradient steps.
  • A decoder-only GPT trained from scratch with embedding size 64 can carry both reconstruction and performance-estimation functions, making the larger encoder-decoder LSTM design unnecessary.
  • The generated feature spaces transfer across downstream model families, so one transformed dataset can serve linear models, trees, SVMs, and neural models without re-running the transformation search.
  • Because inference time is roughly halved on several datasets, the approach is more plausible for online or large-scale feature engineering workloads.
  • The ablations imply that both RL-collected records and gradient-ascent refinement contribute to the gains, so neither stage can be dropped without losing performance.

Reading between the lines

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

  • A direct extension the authors do not explore: the same joint reconstruction-plus-performance embedding could be applied to other discrete search problems, such as neural architecture search, prompt optimization, or program synthesis, wherever sequence generation and a scalar reward can be paired.
  • Because the training records come from a specific RL collector, GPT-FT's ceiling is inherited from the diversity of that collector; if the collector never generates certain useful operation patterns, gradient ascent cannot recover them however smooth the embedding is.
  • The reported efficiency win is partly architectural, but the paper does not separate the cost of collecting records (Step 1) from the cost of the search itself; a full end-to-end cost comparison would need to count the RL data-collection time as well.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper proposes GPT-FT, a four-stage framework for automated feature transformation: (1) collecting transformation records from an RL-based feature generation framework (GRFG), (2) training a small decoder-only GPT with joint sequence-reconstruction (NLL) and performance-estimation (MSE) losses to build a continuous embedding space, (3) performing gradient ascent on the performance head starting from the top-k seed embeddings, and (4) autoregressively reconstructing candidate transformation sequences and selecting the feature space with the highest downstream performance. The authors claim that GPT-FT matches or exceeds state-of-the-art baselines while significantly reducing parameter size and inference time, based on experiments on 15 classification/regression datasets compared against nine baselines, plus ablations and robustness checks.

Significance. If the results are correct, the efficiency contribution is meaningful: Tables 4-5 show roughly a 50% parameter reduction and substantial inference speedups over the LSTM-based MOAT, and the idea of a compact decoder-only transformer with a dual reconstruction/performance head is a reasonable step for continuous AFT search. The paper also provides ablation and sensitivity analyses (Figures 4-6) that help localize the contribution of Steps 1 and 3. However, the central performance claim is undermined by an internal contradiction in the main results table and by the absence of statistical rigor. The paper's significance therefore depends on correcting the experimental reporting and re-framing the claims.

major comments (4)
  1. [Section 4.2, Table 1] The text states that "Table 1 compares GPT-FT with other models on F1-score and 1-RAE metrics, showing GPT-FT outperforms all others across datasets." This is directly contradicted by the Geographical Origin of Music row: DIFER scores 0.632 and GPT-FT scores 0.508, a 0.124 gap in DIFER's favor. Since Table 1 is the central evidence for the headline claim, either the table or the claim is wrong. This must be corrected and the affected conclusions re-evaluated.
  2. [Section 3.4] The assertion "The performance satisfies G(Ē) ≥ G(E)" is a property of gradient ascent on the fitted surrogate G, not a guarantee about actual downstream performance. The real performance after decoding is only empirically verified. The paper itself notes that α∈[0.4,0.9] "fails to generate valid records" and that embedding size 32 yields "occasional invalid records" (Section 4.2), showing the surrogate-to-real transfer is fragile. The authors should present direct evidence (e.g., decoding results over many seeds) that gradient ascent on G reliably improves P, and discuss the invalid-record cases.
  3. [Section 4.2, Tables 1-3] All performance tables report a single run without standard deviations, confidence intervals, or significance tests. Many differences are small (e.g., Heart Disease: GPT-FT 0.867 vs. MOAT 0.866; Weather: 0.980 vs. 0.976). Without repeated trials, the claimed superiority is not statistically supported. In addition, Section 3.5 selects the final feature set as the one yielding the highest downstream performance; if this selection is not also applied to baselines, the comparison is unfair. Report means and variances over multiple runs, and specify the selection protocol for all methods.
  4. [Section 3.2 and Baseline List] Step 1 collects transformation records using GRFG, which is itself one of the baselines. Training GPT-FT on GRFG's RL-collected records and then comparing against GRFG is not an independent evaluation. The authors should clarify what GPT-FT adds beyond GRFG's own search, for example by comparing against a variant trained on records from a different collector or from random transformation sequences, to show that the gains are not inherited from the data collector.
minor comments (6)
  1. [Table 1] The dataset name "Geographical OriginalofMusic" should be "Geographical Origin of Music" (space missing).
  2. [Section 3.3] The NLL loss expression is written as a sum over log probabilities, but the distribution p(γ̂_i | γ_i) is not parameterized. Specify the token-level distribution and how teacher forcing is used during training.
  3. [Section 3.4] The symbol G is used loosely: it is first called "GPT-FT" and then differentiated as ∂G/∂E. Clarify that the gradient is taken with respect to the embedding through the task classifier head δ, not the whole model.
  4. [Section 4.2, Parameter Sensitivity α] "α ∈ [0.4, 0.9] fails to generate valid records" is a strong statement but no quantitative criterion is given. Define "valid records" and report how many seeds/sequences failed in each setting.
  5. [Figure 4 and Ablation] The caption "NoDataPreprocess" is ambiguous: the ablation removes Step 1 (transformation records collection), not generic data preprocessing. Rename for clarity.
  6. [Appendix A.2] The top-42 seed count and α=0.133 are reported after NNI optimization, but no validation/test split for this tuning is described. Clarify whether selection of these hyperparameters used the test set.

Circularity Check

1 steps flagged · score 4.0 of 10

Surrogate gradient-ascent inequality is tautological; main empirical results are self-contained but headline claim contradicts Table 1.

  1. fitted input called prediction [Section 3.4, Gradient-Ascent Search]
    "Starting from an embeddingE, the search updates as ˜E = E + η ∂G ∂E, where ˜E is the refined embedding,η is the step size, andG represents GPT-FT. The performance satisfiesG( ˜E) ≥ G(E)."

    The inequality is not a statement about real downstream performance; it is a first-order Taylor consequence of the gradient-ascent update on the learned surrogate G. Since G is the GPT-FT performance head fitted to the Step-1 records (vi), 'performance' here means G's own estimate, so the claimed improvement is true by construction of the update rule and carries no independent information about actual F1/1-RAE. Presenting this as the mechanism of 'performance enhancement' in Step 3 makes the derivation-level claim reduce to the surrogate's self-consistency rather than to any measured outcome.

full rationale

The derivation chain is mostly self-contained: Step 1's transformation records come from the authors' prior GRFG ([40]) and the main baselines include the authors' own GRFG and MOAT ([41]), but this is ordinary incremental research rather than circularity, since the final F1/1-RAE results are measured on held-out downstream models, not read off the surrogate. The one genuinely circular-by-construction step is Section 3.4: the sentence 'The performance satisfies G(Ẽ) ≥ G(E)' is a Taylor-series consequence of the update Ẽ = E + η∂G/∂E on the fitted performance head G. It states a property of the learned estimator, not of real downstream performance, so as a derivation of 'performance enhancement' it reduces to the surrogate's definition. This raises the score to 4 but does not make the whole paper circular, because the experimental section validates candidate sequences with actual predictive models and selects by measured performance. Separately, the text claims 'GPT-FT outperforms all others across datasets' but Table 1 shows DIFER (0.632) beating GPT-FT (0.508) on Geographical Origin of Music; this is an internal inconsistency and a correctness risk, not a circularity, and does not further raise the circularity score.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

GPT-FT introduces no new physical or ontological entities. The central claim rests on tuned hyperparameters (α, embedding size, layer count, top-k seeds) and on four domain assumptions: the GRFG collector provides representative data, the embedding space is smooth enough for surrogate gradient ascent to transfer to real performance, the evaluation protocol is consistent and held-out, and the joint NLL/MSE training yields valid decodable transformations. None of these is verified with error bars, a described split protocol, or released code. The smoothness assumption is the most fragile: the paper itself documents regions of hyperparameter space, α ≥ 0.4 and embedding size 32, where validity collapses.

free parameters (5)
  • trade-off weight α = 0.133, tuned via NNI after restricting the search range to [0.1, 0.3]
    Balances reconstruction loss Lpre and performance-estimation loss Lcls in L = αLpre + (1-α)Lcls; values above 0.3 failed to generate valid records (Section 4.2, Appendix A.2).
  • embedding size = 64
    Selected from {32, 64, 128, 256, 512, 1024}; size 32 occasionally produced invalid records and sizes 64-1024 performed similarly, so 64 was picked for small model size (Section 4.2).
  • number of embedding-generator layers = 1
    Varied from 1 to 5; differences are minimal, so a single layer was chosen to minimize inference time and model size (Section 4.2).
  • top-k seed count = 42
    The 42 highest-scoring transformation records seed gradient ascent (Appendix A.2); no sensitivity analysis is provided for this choice.
  • learning rate, batch size = 1.31e-5, 16
    Stated in Appendix A.2 without a documented tuning protocol; standard optimizer and batching hyperparameters that affect training convergence and final performance.
assumptions (4)
  • domain assumption GRFG's RL-based framework [40] yields representative, high-quality transformation records with performance scores for every dataset
    All GPT-FT training records come from this collector (Section 3.2); if GRFG's exploration is biased, the embedding space and every downstream search result inherit that bias.
  • domain assumption The learned embedding space is smooth and calibrated, so gradient ascent on the fitted surrogate G transfers to real downstream performance
    Invoked in Section 3.4 where the paper asserts G(Ẽ) ≥ G(E); the paper's own ablations show decoded-sequence validity breaks for α ≥ 0.4 and embedding size 32 (Section 4.2), so this premise is fragile.
  • domain assumption The evaluation protocol (data split, downstream model Q, final feature-space selection) is consistent across record collection, search, and reporting
    Section 3.5 selects the feature space with highest downstream performance and Section 4 reports it, but no split or selection set is described; if selection uses the test data, reported numbers are inflated.
  • domain assumption NLL and MSE losses over the training records produce a usable generative model over valid postfix transformation sequences
    Section 3.3 uses NLL and MSE losses over GRFG records; autoregressive decoding of arbitrary embedding points is not guaranteed to produce valid transformations, as the α and embedding-size ablations show.

how reviews work

0 comments
Cite this review

Pith. "Pith review of GPT-FT: An Efficient Automated Feature Transformation Using GPT for Sequence Reconstruction and Performance Enhancement." pith.science (2026). https://pith.science/paper/2Z4VXX2W

@misc{pith2026250820824,
  author       = {Pith},
  title        = {Pith review of: GPT-FT: An Efficient Automated Feature Transformation Using GPT for Sequence Reconstruction and Performance Enhancement},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2Z4VXX2W}},
  note         = {Machine review of arXiv:2508.20824}
}
read the original abstract

Feature transformation plays a critical role in enhancing machine learning model performance by optimizing data representations. Recent state-of-the-art approaches address this task as a continuous embedding optimization problem, converting discrete search into a learnable process. Although effective, these methods often rely on sequential encoder-decoder structures that cause high computational costs and parameter requirements, limiting scalability and efficiency. To address these limitations, we propose a novel framework that accomplishes automated feature transformation through four steps: transformation records collection, embedding space construction with a revised Generative Pre-trained Transformer (GPT) model, gradient-ascent search, and autoregressive reconstruction. In our approach, the revised GPT model serves two primary functions: (a) feature transformation sequence reconstruction and (b) model performance estimation and enhancement for downstream tasks by constructing the embedding space. Such a multi-objective optimization framework reduces parameter size and accelerates transformation processes. Experimental results on benchmark datasets show that the proposed framework matches or exceeds baseline performance, with significant gains in computational efficiency. This work highlights the potential of transformer-based architectures for scalable, high-performance automated feature transformation.

Figures

Figures reproduced from arXiv: 2508.20824 by the authors.

Figure 1
Figure 1. An overview of our model framework GPT-FT. [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Comparison of parameter size across different datasets. We normalize the [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Comparison of inference times across different datasets. We normalize the [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Comparison of performance across datasets, considering (a) data pre [PITH_FULL_IMAGE:figures/full_fig_p011_4.png]
Figure 5
Figure 5. Figure 5: Trends in Lcls and Lpre during training for α = 0.1, 0.2, . . . , 0.9, demon￾strating consistent behavior across different values of α. 0 0.5 1 1.5 2 2.5 3 3.5 4 0 500 1000 1500 2000 Loss Epoch 0.07 0.17 0.27 0.37 0.47 0.57 0.67 65 115 165 215 265 5 2 Layer=134 (a) Inc…
Figure 6
Figure 6. Figure 6: Training trends in GPT-FT with varying embedding generator layer num [PITH_FULL_IMAGE:figures/full_fig_p012_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

49 extracted references · 42 canonical work pages

  1. [1]

    Journal of machine Learning research 3(Jan), 993–1022 (2003)

    Blei, D.M., Ng, A.Y., Jordan, M.I.: Latent dirichlet allocation. Journal of machine Learning research 3(Jan), 993–1022 (2003)

    work page 2003

  2. [2]

    UCI Machine Learning Repository (2010), DOI: https://doi.org/10.24432/C5H30K

    Charytanowicz, M., Niewczas, J., Kulczycki, P., Kowalski, P., Lukasik, S.: Seeds. UCI Machine Learning Repository (2010), DOI: https://doi.org/10.24432/C5H30K

    doi:10.24432/c5h30k 2010

  3. [3]

    In: 2019 IEEE International Conference on Data Mining (ICDM)

    Chen, X., Lin, Q., Luo, C., Li, X., Zhang, H., Xu, Y., Dang, Y., Sui, K., Zhang, X., Qiao, B., et al.: Neural feature search: A neural architecture for automated feature engineering. In: 2019 IEEE International Conference on Data Mining (ICDM). pp. 71–80. IEEE (2019)

    work page 2019

  4. [4]

    ACM SIGKDD Explorations Newsletter22(2), 35–50 (2021)

    Chen, Y.W., Song, Q., Hu, X.: Techniques for automated machine learning. ACM SIGKDD Explorations Newsletter22(2), 35–50 (2021)

    work page 2021

  5. [5]

    UCI Machine Learning Repository (2009), DOI: https://doi.org/10.24432/C56S3T

    Cortez, P., Cerdeira, A., Almeida, F., Matos, T., Reis, J.: Wine Quality. UCI Machine Learning Repository (2009), DOI: https://doi.org/10.24432/C56S3T

    doi:10.24432/c56s3t 2009

  6. [6]

    UCI Machine Learning Reposi- tory (2009), DOI: https://doi.org/10.24432/C5GC82

    Dias, D., Peres, S., Bscaro, H.: Libras Movement. UCI Machine Learning Reposi- tory (2009), DOI: https://doi.org/10.24432/C5GC82

    work page doi:10.24432/c5gc82 2009

  7. [7]

    Infor- mation Sciences 189, 176–190 (2012)

    Dor, O., Reich, Y.: Strengthening learning algorithms by feature discovery. Infor- mation Sciences 189, 176–190 (2012)

    work page 2012

  8. [8]

    The Journal of Machine Learning Research20(1), 1997–2017 (2019)

    Elsken, T., Metzen, J.H., Hutter, F.: Neural architecture search: A survey. The Journal of Machine Learning Research20(1), 1997–2017 (2019)

    work page 1997

Show all 49 references
  1. [9]

    The Journal of Supercomputing 80(18), 26394–26434 (2024)

    Gao, Y., Quan, G., Homsi, S., Wen, W., Wang, L.: Secure and efficient general matrix multiplication on cloud using homomorphic encryption. The Journal of Supercomputing 80(18), 26394–26434 (2024)

    2024

  2. [10]

    Knowledge- Based Systems 212, 106622 (2021)

    He, X., Zhao, K., Chu, X.: Automl: A survey of the state-of-the-art. Knowledge- Based Systems 212, 106622 (2021)

    2021

  3. [11]

    arXiv preprint arXiv:1901.07329 (2019)

    Horn, F., Pack, R., Rieger, M.: The autofeat python library for automated feature engineering and selection. arXiv preprint arXiv:1901.07329 (2019)

    1901 arXiv

  4. [12]

    In: Machine Learning and Knowledge Discovery in Databases: International Workshops of ECML PKDD 2019, Würzburg, Germany, September 16–20, 2019, Proceedings, Part I

    Horn, F., Pack, R., Rieger, M.: The autofeat python library for automated fea- ture engineering and selection. In: Machine Learning and Knowledge Discovery in Databases: International Workshops of ECML PKDD 2019, Würzburg, Germany, September 16–20, 2019, Proceedings, Part I. p...

    2019

  5. [13]

    Howard, J.: Kaggle dataset download.https://www.kaggle.com/datasets (2022), [EB/OL]

    2022

  6. [14]

    J, M.: Weather dataset (nd), https://www.kaggle.com/datasets/muthuj7/ weather-dataset/data, accessed: 2024-11-23

    2024

  7. [15]

    UCI Machine Learning Repository (1989), DOI: https://doi.org/10.24432/C52P4X

    Janosi, A., Steinbrunn, W., Pfisterer, M., Detrano, R.: Heart Disease. UCI Machine Learning Repository (1989), DOI: https://doi.org/10.24432/C52P4X

    1989 doi

  8. [16]

    Jour- nal of Statistics Education4 (1996), http://www.amstat.org/publications/jse/ v4n1/datasets.johnson.html, bodyfat data retrieved from the American Statis- tical Association

    Johnson, W.: Fitting percentage of body fat to simple body measurements. Jour- nal of Statistics Education4 (1996), http://www.amstat.org/publications/jse/ v4n1/datasets.johnson.html, bodyfat data retrieved from the American Statis- tical Association. An expanded version is in...

    1996

  9. [17]

    Kaggle: Titanic - machine learning from disaster (nd),https://www.kaggle.com/ c/titanic

  10. [18]

    In: 2015 IEEE international conference on data science and advanced analytics (DSAA)

    Kanter, J.M., Veeramachaneni, K.: Deep feature synthesis: Towards automating data science endeavors. In: 2015 IEEE international conference on data science and advanced analytics (DSAA). pp. 1–10. IEEE (2015)

    2015

  11. [19]

    ACM Computing Surveys (CSUR) 54(8), 1–36 (2021)

    Karmaker, S.K., Hassan, M.M., Smith, M.J., Xu, L., Zhai, C., Veeramachaneni, K.: Automl to date and beyond: Challenges and opportunities. ACM Computing Surveys (CSUR) 54(8), 1–36 (2021)

    2021

  12. [20]

    In: 2016 IEEE 16th International Conference on Data Mining (ICDM)

    Katz, G., Shin, E.C.R., Song, D.: Explorekit: Automatic feature generation and selection. In: 2016 IEEE 16th International Conference on Data Mining (ICDM). pp. 979–984. IEEE (2016)

    2016

  13. [21]

    Kelly, M., Longjohn, R., Nottingham, K.: The uci machine learning repository, https://archive.ics.uci.edu, accessed: 2024-10-30

    2024

  14. [22]

    Transformation10(10), 10 (2016)

    Khurana, U., Nargesian, F., Samulowitz, H., Khalil, E., Turaga, D.: Automating feature engineering. Transformation10(10), 10 (2016)

    2016

  15. [23]

    In: Proceedings of the AAAI Conference on Artificial Intelligence

    Khurana, U., Samulowitz, H., Turaga, D.: Feature engineering for predictive mod- eling using reinforcement learning. In: Proceedings of the AAAI Conference on Artificial Intelligence. vol. 32 (2018)

    2018

  16. [24]

    In: 2016 IEEE 16th international con- ference on data mining workshops (ICDMW)

    Khurana, U., Turaga, D., Samulowitz, H., Parthasrathy, S.: Cognito: Automated feature engineering for supervised learning. In: 2016 IEEE 16th international con- ference on data mining workshops (ICDMW). pp. 1304–1307. IEEE (2016)

    2016

  17. [25]

    IEEE Transactions on Electronics Packaging Manufacturing24(3), 214–221 (2001)

    Kusiak, A.: Feature transformation methods in data mining. IEEE Transactions on Electronics Packaging Manufacturing24(3), 214–221 (2001)

    2001

  18. [26]

    arXiv preprint arXiv:1706.00327 (2017)

    Lam, H.T., Thiebaut, J.M., Sinn, M., Chen, B., Mai, T., Alkan, O.: One button machine for automating feature engineering in relational databases. arXiv preprint arXiv:1706.00327 (2017)

    2017 arXiv

  19. [27]

    In: Proceedings of the 30th ACM International Conference on Information & Knowledge Management

    Li, Y., Wang, Z., Xie, Y., Ding, B., Zeng, K., Zhang, C.: Automl: From methodol- ogy to application. In: Proceedings of the 30th ACM International Conference on Information & Knowledge Management. pp. 4853–4856 (2021)

    2021

  20. [28]

    UCI Machine Learning Repository (1999), DOI: https://doi.org/10.24432/C59W2D

    Lim, T.S.: Contraceptive Method Choice. UCI Machine Learning Repository (1999), DOI: https://doi.org/10.24432/C59W2D

    1999 doi

  21. [29]

    In: Pro- ceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V

    Liu, R., Xie, R., Yao, Z., Fu, Y., Wang, D.: Continuous optimization for feature selection with permutation-invariant embedding and policy-guided search. In: Pro- ceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V. 2. pp. 1857–1866 (2025)

    2025

  22. [30]

    In: 2024 IEEE International Conference on Big Data (BigData)

    Liu, S., Gao, Y., Zhai, S., Wang, L.: Stylerec: A benchmark dataset for prompt recovery in writing style transformation. In: 2024 IEEE International Conference on Big Data (BigData). pp. 1678–1685. IEEE (2024)

    2024

  23. [31]

    Massaron, L.: Amazon employee access challenge (nd),https://www.kaggle.com/ datasets/lucamassaron/amazon-employee-access-challenge, accessed: 2024- 11-23

    2024

  24. [32]

    Microsoft:NeuralNetworkIntelligence(12021), https://github.com/microsoft/ nni

  25. [33]

    In: Proceedings of the genetic and evolutionary computation conference 2016

    Olson, R.S., Bartley, N., Urbanowicz, R.J., Moore, J.H.: Evaluation of a tree- based pipeline optimization tool for automating data science. In: Proceedings of the genetic and evolutionary computation conference 2016. pp. 485–492 (2016)

    2016

  26. [34]

    Public: Openml dataset download.https://www.openml.org (2022), [EB/OL]

    2022

  27. [35]

    Radford, A.: Improving language understanding by generative pre-training (2018)

    2018

  28. [36]

    Ramadwiyantara, R.: Jasmine dataset (nd),https://www.kaggle.com/datasets/ ramadwiyantara/jasmine-dataset, accessed: 2024-11-23

    2024

  29. [37]

    Tecator: Tecator dataset: Food spectroscopy analysis (nd), https://rdrr.io/ cran/caret/man/tecator.html, available from the R package ‘caret‘ or online repositories

  30. [38]

    Memetic Computing8(1), 3–15 (2016)

    Tran, B., Xue, B., Zhang, M.: Genetic programming for feature construction and selection in classification on high-dimensional data. Memetic Computing8(1), 3–15 (2016)

    2016

  31. [39]

    Pattern Recognition93, 404–417 (2019)

    Tran, B., Xue, B., Zhang, M.: Genetic programming for multiple-feature construc- tion on high-dimensional classification. Pattern Recognition93, 404–417 (2019)

    2019

  32. [40]

    In: Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining

    Wang, D., Fu, Y., Liu, K., Li, X., Solihin, Y.: Group-wise reinforcement feature generation for optimal and explainable representation space reconstruction. In: Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. pp. 1826–1834 (2022)

    2022

  33. [41]

    Advances in Neural Information Processing Systems36, 43563–43578 (2023)

    Wang, D., Xiao, M., Wu, M., Zhou, Y., Fu, Y., et al.: Reinforcement-enhanced autoregressive feature transformation: Gradient-steered search in continuous space for postfix expressions. Advances in Neural Information Processing Systems36, 43563–43578 (2023)

    2023

  34. [42]

    Wu, Y., Huang, Y., Chen, Z., Yao, Z., Fu, Y., Liu, K., Luo, X., Wang, D.: Iterative featurespaceoptimizationthroughincrementaladaptiveevaluation.arXivpreprint arXiv:2501.14889 (2025)

    2025 arXiv

  35. [43]

    arXiv preprint arXiv:2306.16893 (2023)

    Xiao, M., Wang, D., Wu, M., Liu, K., Xiong, H., Zhou, Y., Fu, Y.: Traceable group-wise self-optimizing feature transformation learning: A dual optimization perspective. arXiv preprint arXiv:2306.16893 (2023)

    2023 arXiv

  36. [44]

    In: Pro- ceedings of the 2023 SIAM International Conference on Data Mining (SDM)

    Xiao, M., Wang, D., Wu, M., Qiao, Z., Wang, P., Liu, K., Zhou, Y., Fu, Y.: Trace- able automatic feature transformation via cascading actor-critic agents. In: Pro- ceedings of the 2023 SIAM International Conference on Data Mining (SDM). pp. 775–783. SIAM (2023)

    2023

  37. [45]

    UCI Machine Learning Repository (2008), DOI: https://doi.org/10.24432/C5NG6W

    Zhang, K., Fan, W., Yuan, X.: Ozone Level Detection. UCI Machine Learning Repository (2008), DOI: https://doi.org/10.24432/C5NG6W

    2008 doi

  38. [46]

    UCI Machine Learning Repository (2014), DOI: https://doi.org/10.24432/C5VK5D

    Zhou, F.: Geographical Origin of Music. UCI Machine Learning Repository (2014), DOI: https://doi.org/10.24432/C5VK5D

    2014 doi

  39. [47]

    Zhu, G., Jiang, S., Guo, X., Yuan, C., Huang, Y.: Evolutionary automated feature engineering. In: PRICAI 2022: Trends in Artificial Intelligence: 19th Pacific Rim International Conference on Artificial Intelligence, PRICAI 2022, Shanghai, China, November 10–13, 2022, Proceedin...

    2022

  40. [48]

    In: International Conference on Automated Machine Learning

    Zhu, G., Xu, Z., Yuan, C., Huang, Y.: Difer: differentiable automated feature engi- neering. In: International Conference on Automated Machine Learning. pp. 17–1. PMLR (2022)

    2022

  41. [49]

    Zwitter, M., Soklic, M.: Lymphography. UCI Machine Learning Repository (1988), DOI: https://doi.org/10.24432/C54598 A Experiment A.1 Experiment Platform Information All experiments were conducted on the Ubuntu 20.04.6 LTS operating system, Intel(R) Xeon(R) Silver 4114 CPU, and...

    1988 doi

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.