Jump to content

Talk:Resampling (statistics)

Page contents not supported in other languages.
From Wikipedia, the free encyclopedia

What kind of resampling is RANSAC

[edit]

I saw RANSAC once described in a as "RANSAC [...] being a form of crossvalidation". Does somebody have one or more reliable citations for this? Is this correct? 2A01:C22:CD3E:6B00:A58E:E4A7:CE50:ADB2 (talk) 19:21, 15 October 2021 (UTC)[reply]

This paper on RANSAC highlights that subsampling is a key ingredient: http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.106.3035&rep=rep1&type=pdf (Random Sample Consensus (RANSAC) H. Cantzler ). This book says: "Perhaps the most famous example of the cross validation approach is RANSAC" on page 92 https://www.google.de/books/edition/Robotics_Research/sofbBwAAQBAJ?hl=de&gbpv=1&dq=ransac%20cross%20validation&pg=PA92&printsec=frontcover&bsq=ransac%20cross%20validation ISBN:9781447107651 . Apparently they mean Cross-validation_(statistics)#Repeated random sub-sampling validation. I will restructure the article to reflect that subsampling can be viewed as Cross validation method. Biggerj1 (talk) 19:43, 15 October 2021 (UTC)[reply]

 Done

Subsampling

[edit]

I think it might we worthwhile to add a brief section on Subsampling (good reference is Politis, Romano, Wolf 1999). Bootstrap frequently fails if convergence is not root n or if the CLT is not applicable, while Subsampling still works in these cases. Oragonof (talk) 15:22, 14 December 2022 (UTC)[reply]