{"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/why-comparing-single-performance-scores-does","title":"Why Comparing Single Performance Scores Does Not Allow to Draw Conclusions About Machine Learning Approaches","arxiv_id":"1803.09578","date":"2018-03-26","proceeding":null,"authors":["Nils Reimers","Iryna Gurevych"],"abstract":"Developing state-of-the-art approaches for specific tasks is a major driving\nforce in our research community. Depending on the prestige of the task,\npublishing it can come along with a lot of visibility. The question arises how\nreliable are our evaluation methodologies to compare approaches?\n  One common methodology to identify the state-of-the-art is to partition data\ninto a train, a development and a test set. Researchers can train and tune\ntheir approach on some part of the dataset and then select the model that\nworked best on the development set for a final evaluation on unseen test data.\nTest scores from different approaches are compared, and performance differences\nare tested for statistical significance.\n  In this publication, we show that there is a high risk that a statistical\nsignificance in this type of evaluation is not due to a superior learning\napproach. Instead, there is a high risk that the difference is due to chance.\nFor example for the CoNLL 2003 NER dataset we observed in up to 26% of the\ncases type I errors (false positives) with a threshold of p < 0.05, i.e.,\nfalsely concluding a statistically significant difference between two identical\napproaches.\n  We prove that this evaluation setup is unsuitable to compare learning\napproaches. We formalize alternative evaluation setups based on score\ndistributions.","url_abs":"http://arxiv.org/abs/1803.09578v1","url_pdf":"http://arxiv.org/pdf/1803.09578v1.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":"why-comparing-single-performance-scores-does","repo_url":"https://github.com/UKPLab/emnlp2017-bilstm-cnn-crf","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok","spdx":"Apache-2.0"}}],"tasks":[{"task_slug":"machine-learning","task_name":"BIG-bench Machine Learning"},{"task_slug":"cg","task_name":"NER"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1803.09578","mcp":{"get_harvested_code_for_paper":{"arxiv_id":"1803.09578"}},"developers":"https://syntology.ai/developers","read_at":"2026-09-24T18:15:14+00:00","read_at_is":"when the build read Syntology's graph, not when any sample ran","claim":"Per-sample execution status on synthesized fixtures; not a correctness claim about the paper. Samples come from repositories linked to the paper, official or community; repo_kind says which.","repos":[{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/UKPLab/emnlp2017-bilstm-cnn-crf","reach":{"status":"ok","spdx":"Apache-2.0"}}],"summary":{"unverified":6},"by_repo_kind":{"official":{"samples":6,"ran":0,"repositories":1}},"repo_kind_vocabulary":{"official":"The archive marks this repository official for the paper","named_in_paper":"The archive records that the paper mentions this repository; it is not marked official","listed":"In the archive's code links for this paper, not marked official and not recorded as mentioned in the paper","found_in_text":"Syntology found this repository in the paper's own text; whether it is the authors' implementation is not asserted","community":"Not in the archive's code links for this paper; a community repository Syntology harvested"},"n_pointer_only_for_licence":0,"samples":[{"code_sha256_prefix":"2561978b0fd1e5d3","entry":"compute_f1","repo":"UKPLab/emnlp2017-bilstm-cnn-crf","repo_kind":"official","path":"util/BIOF1Validation.py","file_url":"https://github.com/UKPLab/emnlp2017-bilstm-cnn-crf/blob/HEAD/util/BIOF1Validation.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"2561978b0fd1e5d3"}},{"code_sha256_prefix":"1b66320c71babcd2","entry":"compute_f1_token_basis","repo":"UKPLab/emnlp2017-bilstm-cnn-crf","repo_kind":"official","path":"util/BIOF1Validation.py","file_url":"https://github.com/UKPLab/emnlp2017-bilstm-cnn-crf/blob/HEAD/util/BIOF1Validation.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"1b66320c71babcd2"}},{"code_sha256_prefix":"6e5cd3fa57bb3c06","entry":"compute_precision_token_basis","repo":"UKPLab/emnlp2017-bilstm-cnn-crf","repo_kind":"official","path":"util/BIOF1Validation.py","file_url":"https://github.com/UKPLab/emnlp2017-bilstm-cnn-crf/blob/HEAD/util/BIOF1Validation.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"6e5cd3fa57bb3c06"}},{"code_sha256_prefix":"6f5b259e1a881655","entry":"path_energy","repo":"UKPLab/emnlp2017-bilstm-cnn-crf","repo_kind":"official","path":"neuralnets/keraslayers/ChainCRF.py","file_url":"https://github.com/UKPLab/emnlp2017-bilstm-cnn-crf/blob/HEAD/neuralnets/keraslayers/ChainCRF.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"6f5b259e1a881655"}},{"code_sha256_prefix":"e13517b914ec8cfe","entry":"path_energy0","repo":"UKPLab/emnlp2017-bilstm-cnn-crf","repo_kind":"official","path":"neuralnets/keraslayers/ChainCRF.py","file_url":"https://github.com/UKPLab/emnlp2017-bilstm-cnn-crf/blob/HEAD/neuralnets/keraslayers/ChainCRF.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"e13517b914ec8cfe"}},{"code_sha256_prefix":"29092b6b1ee0e8b7","entry":"sparse_chain_crf_loss","repo":"UKPLab/emnlp2017-bilstm-cnn-crf","repo_kind":"official","path":"neuralnets/keraslayers/ChainCRF.py","file_url":"https://github.com/UKPLab/emnlp2017-bilstm-cnn-crf/blob/HEAD/neuralnets/keraslayers/ChainCRF.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"29092b6b1ee0e8b7"}}]},"arxiv_metadata":null,"syntology_extracted_results":null}