Home > Research > Publications & Outputs > Using J-K-fold Cross Validation to Reduce Varia...

Electronic data

  • coling2018

    Accepted author manuscript, 453 KB, PDF document

    Available under license: CC BY: Creative Commons Attribution 4.0 International License

Links

View graph of relations

Using J-K-fold Cross Validation to Reduce Variance When Tuning NLP Models

Research output: Contribution in Book/Report/Proceedings - With ISBN/ISSNConference contribution/Paperpeer-review

E-pub ahead of print

Standard

Using J-K-fold Cross Validation to Reduce Variance When Tuning NLP Models. / Moss, Henry; Leslie, David Stuart; Rayson, Paul Edward.
Proceedings of COLING 2018. Association for Computational Linguistics (ACL Anthology), 2018. p. 2978–2989 (Proceedings of COLING 2018).

Research output: Contribution in Book/Report/Proceedings - With ISBN/ISSNConference contribution/Paperpeer-review

Harvard

Moss, H, Leslie, DS & Rayson, PE 2018, Using J-K-fold Cross Validation to Reduce Variance When Tuning NLP Models. in Proceedings of COLING 2018. Proceedings of COLING 2018, Association for Computational Linguistics (ACL Anthology), pp. 2978–2989, Conference on Computational Linguistics, Santa Fe, New Mexico, United States, 20/08/18. <https://aclanthology.org/C18-1252>

APA

Moss, H., Leslie, D. S., & Rayson, P. E. (2018). Using J-K-fold Cross Validation to Reduce Variance When Tuning NLP Models. In Proceedings of COLING 2018 (pp. 2978–2989). (Proceedings of COLING 2018). Association for Computational Linguistics (ACL Anthology). Advance online publication. https://aclanthology.org/C18-1252

Vancouver

Moss H, Leslie DS, Rayson PE. Using J-K-fold Cross Validation to Reduce Variance When Tuning NLP Models. In Proceedings of COLING 2018. Association for Computational Linguistics (ACL Anthology). 2018. p. 2978–2989. (Proceedings of COLING 2018). Epub 2018 Jun.

Author

Moss, Henry ; Leslie, David Stuart ; Rayson, Paul Edward. / Using J-K-fold Cross Validation to Reduce Variance When Tuning NLP Models. Proceedings of COLING 2018. Association for Computational Linguistics (ACL Anthology), 2018. pp. 2978–2989 (Proceedings of COLING 2018).

Bibtex

@inproceedings{5e32f30ed1d745a98b4b8eab70c601bf,
title = "Using J-K-fold Cross Validation to Reduce Variance When Tuning NLP Models",
abstract = "K-fold cross validation (CV) is a popular method for estimating the true performance of machine learning models, allowing model selection and parameter tuning. However, the very process of CV requires random partitioning of the data and so our performance estimates are in fact stochastic, with variability that can be substantial for natural language processing tasks. We demonstrate that these unstable estimates cannot be relied upon for effective parameter tuning. The resulting tuned parameters are highly sensitive to how our data is partitioned, meaning that we often select sub-optimal parameter choices and have serious reproducibility issues.Instead, we propose to use the less variable J-K-fold CV, in which J independent K-fold cross validations are used to assess performance. Our main contributions are extending J-K-fold CV from performance estimation to parameter tuning and investigating how to choose J and K. We argue that variability is more important than bias for effective tuning and so advocate lower choices of K than are typically seen in the NLP literature, instead use the saved computation to increase J. To demonstrate the generality of our recommendations we investigate a wide range of case-studies: sentiment classification (both general and target-specific), part-of-speech tagging and document classification.",
author = "Henry Moss and Leslie, {David Stuart} and Rayson, {Paul Edward}",
note = "COLING 2018. Code available at: https://github.com/henrymoss/COLING2018; Conference on Computational Linguistics, COLING ; Conference date: 20-08-2018 Through 26-08-2018",
year = "2018",
month = jun,
language = "English",
series = "Proceedings of COLING 2018",
publisher = "Association for Computational Linguistics (ACL Anthology)",
pages = "2978–2989",
booktitle = "Proceedings of COLING 2018",
url = "https://coling2018.org/",

}

RIS

TY - GEN

T1 - Using J-K-fold Cross Validation to Reduce Variance When Tuning NLP Models

AU - Moss, Henry

AU - Leslie, David Stuart

AU - Rayson, Paul Edward

N1 - Conference code: 27

PY - 2018/6

Y1 - 2018/6

N2 - K-fold cross validation (CV) is a popular method for estimating the true performance of machine learning models, allowing model selection and parameter tuning. However, the very process of CV requires random partitioning of the data and so our performance estimates are in fact stochastic, with variability that can be substantial for natural language processing tasks. We demonstrate that these unstable estimates cannot be relied upon for effective parameter tuning. The resulting tuned parameters are highly sensitive to how our data is partitioned, meaning that we often select sub-optimal parameter choices and have serious reproducibility issues.Instead, we propose to use the less variable J-K-fold CV, in which J independent K-fold cross validations are used to assess performance. Our main contributions are extending J-K-fold CV from performance estimation to parameter tuning and investigating how to choose J and K. We argue that variability is more important than bias for effective tuning and so advocate lower choices of K than are typically seen in the NLP literature, instead use the saved computation to increase J. To demonstrate the generality of our recommendations we investigate a wide range of case-studies: sentiment classification (both general and target-specific), part-of-speech tagging and document classification.

AB - K-fold cross validation (CV) is a popular method for estimating the true performance of machine learning models, allowing model selection and parameter tuning. However, the very process of CV requires random partitioning of the data and so our performance estimates are in fact stochastic, with variability that can be substantial for natural language processing tasks. We demonstrate that these unstable estimates cannot be relied upon for effective parameter tuning. The resulting tuned parameters are highly sensitive to how our data is partitioned, meaning that we often select sub-optimal parameter choices and have serious reproducibility issues.Instead, we propose to use the less variable J-K-fold CV, in which J independent K-fold cross validations are used to assess performance. Our main contributions are extending J-K-fold CV from performance estimation to parameter tuning and investigating how to choose J and K. We argue that variability is more important than bias for effective tuning and so advocate lower choices of K than are typically seen in the NLP literature, instead use the saved computation to increase J. To demonstrate the generality of our recommendations we investigate a wide range of case-studies: sentiment classification (both general and target-specific), part-of-speech tagging and document classification.

M3 - Conference contribution/Paper

T3 - Proceedings of COLING 2018

SP - 2978

EP - 2989

BT - Proceedings of COLING 2018

PB - Association for Computational Linguistics (ACL Anthology)

T2 - Conference on Computational Linguistics

Y2 - 20 August 2018 through 26 August 2018

ER -