{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/using-j-k-fold-cross-validation-to-reduce","title":"Using J-K fold Cross Validation to Reduce Variance When Tuning NLP Models","arxiv_id":"1806.07139","date":"2018-06-19","proceeding":null,"authors":["Henry B. Moss","David S. Leslie","Paul Rayson"],"abstract":"K-fold cross validation (CV) is a popular method for estimating the true\nperformance of machine learning models, allowing model selection and parameter\ntuning. However, the very process of CV requires random partitioning of the\ndata and so our performance estimates are in fact stochastic, with variability\nthat can be substantial for natural language processing tasks. We demonstrate\nthat these unstable estimates cannot be relied upon for effective parameter\ntuning. The resulting tuned parameters are highly sensitive to how our data is\npartitioned, meaning that we often select sub-optimal parameter choices and\nhave serious reproducibility issues.\n  Instead, we propose to use the less variable J-K-fold CV, in which J\nindependent K-fold cross validations are used to assess performance. Our main\ncontributions are extending J-K-fold CV from performance estimation to\nparameter tuning and investigating how to choose J and K. We argue that\nvariability is more important than bias for effective tuning and so advocate\nlower choices of K than are typically seen in the NLP literature, instead use\nthe saved computation to increase J. To demonstrate the generality of our\nrecommendations we investigate a wide range of case-studies: sentiment\nclassification (both general and target-specific), part-of-speech tagging and\ndocument classification.","url_abs":"http://arxiv.org/abs/1806.07139v1","url_pdf":"http://arxiv.org/pdf/1806.07139v1.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"using-j-k-fold-cross-validation-to-reduce","repo_url":"https://github.com/henrymoss/COLING2018","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"none","reach":null}],"tasks":[{"task_slug":"document-classification","task_name":"Document Classification"},{"task_slug":"classification","task_name":"General Classification"},{"task_slug":"model-selection","task_name":"Model Selection"},{"task_slug":"part-of-speech-tagging","task_name":"Part-Of-Speech Tagging"},{"task_slug":"sentiment-analysis","task_name":"Sentiment Analysis"},{"task_slug":"sentiment-classification","task_name":"Sentiment Classification"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":"https://syntology.ai/paper/1806.07139","atlas_url":"https://app.syntology.ai/?focus=1806.07139","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}