{"artifacts":[{"compression":"none","distribution":"upstream-only","downloads":[{"kind":"upstream","url":"https://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets/binary/madelon"}],"format":"libsvm","name":"train","sha256":"4801c452441772b47674fa8deb3724b1cb897eff49ccd8d47078367080c05b54","size":7788995},{"compression":"none","distribution":"upstream-only","downloads":[{"kind":"upstream","url":"https://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets/binary/madelon.t"}],"format":"libsvm","name":"test","sha256":"ae60251b63a46c6ace348b77749e79e25d4098d629d1dead2852c3d4daf1f8ea","size":2336699}],"description":"Artificial binary classification data from the NIPS 2003 feature selection challenge, distributed by LIBSVM with 500 features. The predefined test artifact contains the original challenge validation set, not its unlabeled test set.","license":{"evidence":"UCI licenses the original Madelon dataset under CC BY 4.0: https://archive.ics.uci.edu/dataset/171/madelon. LIBSVM identifies that source and documents the split selection but states no license terms for its converted artifacts: https://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets/binary.html#madelon. The applicability of the original license to these distributed artifacts has not been established.","status":"unknown"},"modality":"tabular","name":"madelon","preprocessing":{"datamonger":{"evidence":["spec/decoders/libsvm-v2.md"],"notes":"LIBSVM v2 omits explicitly stored zeros without changing feature values. Datamonger applies no statistical preprocessing.","status":"none"},"upstream":{"evidence":["https://archive.ics.uci.edu/dataset/171/madelon","https://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets/binary.html#madelon"],"notes":"The exact numerical transformations used to prepare LIBSVM's feature values have not been established.","status":"documented","steps":[{"description":"Generate clusters on a five-dimensional hypercube, add 15 linear combinations to the five informative features and 480 distractor features, and randomize feature and observation order.","kind":"synthetic-generation"},{"description":"LIBSVM uses the original 600-row validation set as its test set because labels for the original 1800-row test set are unavailable.","kind":"split-selection"}]}},"provenance":{"authors":["Isabelle Guyon"],"citations":[{"citation":"Guyon, I. (2004). Madelon [Dataset]. UCI Machine Learning Repository. https://doi.org/10.24432/C5602H.","note":"Original dataset and creator as cited by UCI.","role":"dataset"},{"citation":"Isabelle Guyon, Steve Gunn, Asa Ben Hur, and Gideon Dror. Result analysis of the NIPS 2003 feature selection challenge. Advances in Neural Information Processing Systems, volume 17, 2005.","note":"Reference IG05a associated with Madelon by LIBSVM: https://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets/ref.html#IG05a.","role":"paper"}],"landing_page":"https://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets/binary.html#madelon","provider":"LIBSVM","retrieved_at":"2026-10-26","upstream_name":"madelon"},"representation":{"decoder":"libsvm-split","decoder_version":2,"expect":{"components":[{"columns":500,"kind":"sparse_matrix","name":"train_features","rows":2000,"type":"float64"},{"kind":"vector","length":2000,"name":"train_response","type":"int64"},{"columns":500,"kind":"sparse_matrix","name":"test_features","rows":600,"type":"float64"},{"kind":"vector","length":600,"name":"test_response","type":"int64"}],"verification":[{"algorithm":"sha256","canonical_form":1,"digest":"997dff34ede32628996d5944cec46a39803bdda495ad8b7830e6e5d372d3a125"}]},"inputs":{"test":"test","train":"train"},"options":{"duplicate_features":"error","feature_count":500,"index_base":1,"label_type":"int64","row_order":"source","target_name":"response"}},"schema_version":5,"source":"libsvm","tasks":[{"name":"default","splits":{"test":{"features":"test_features","target":"test_response"},"train":{"features":"train_features","target":"train_response"}},"type":"classification"}],"title":"Madelon","version":"1"}