Talk:Segmented regression

Latest comment: 8 years ago by PapaNappa in topic Merger of Time-series segmentation

Improving this article

edit

Well over half the material on this page does not belong here, and instead belongs on the appropriate pages such as linear regression. This is a wiki, so if there is a term that needs a definition, the appropriate response is to wikify it (i.e. add the link to the other page) and then ensure that the other page has a clear definition. As such I have deleted a lot of the material and will continue to delete more. Also, I think the chatty tone, which may be appropriate for a textbook, is not appropriate for this article, especially for when the reader is pointed to references. It comes across as highly POV. I think that the appropriate way to do this is to cite in-line references when relevant and then have a list of general references at the end of the article. In that place, people can add a short (and as neutral as possible) description of what that reference is and what audience it is appropriate for. I think it's great to refer the reader to outside sources but the current format of doing so makes this article come across pretty poorly. Cazort (talk) 15:39, 15 December 2007 (UTC)Reply

I agree with you that we should not merge the page. My comments were more about the fact that this page contained a lot of material that was just about interpreting linear regression in general, stuff that was not specific to segmented regression. Cazort (talk) 22:23, 24 December 2007 (UTC)Reply
No doubt that the page needs improvement. However, the relative amount of material about intepreting linear regression in general does, in my view, not seem excessive. In the context of segmented regression, references to linear regression are necessary, at least to a certain extent. Could you help to find the optimal balance? R.J.Oosterbaan (talk) 00:16, 27 December 2007 (UTC)Reply
I have given the page an overhaul accomodating Cazort´s observations. R.J.Oosterbaan (talk) 17:36, 28 December 2007 (UTC)Reply
Thanks! I think this is a big improvement! Cazort (talk) 23:11, 30 December 2007 (UTC)Reply

References?

edit

I am going to recheck the refs for this article and look to find better ones. Of the ones currently here, only the last one seems remotely general enough to be relevant here (the others perhaps useful in the respective soil etc subject areas. Perhaps), and all of them seem from one particular researcher's website. While perhaps not technically a violation of WP:EL, there are certainly far better references for this topic. Books and papers from e.g., Seber & Wild, Fuller, Wahba, Davies, Muggeo, Friedman, Bacon & Watts, etc., come to mind. Baccyak4H (Yak!) 18:37, 22 January 2008 (UTC)Reply

Looking forward to the improvements. R.J.Oosterbaan (talk) 22:38, 23 January 2008 (UTC)Reply

Merger of Time-series segmentation

edit

I do not think that Time-series segmentation should be merged into this article. Time-series segmentation is answering the question: "Given a time-series, what labels can I assign to the disjoint time-intervals?". For instance in the referenced Speaker diarisation, every audio sample is assigned to one speaker. Time-series segmentation always has some model of a dynamic system evolving over time. Regression, on the other hand, tries to find a suitable model of measurement y given input x. One example is Linear regression, e.g. finding a and b such that all  . Regression does not even assume that any of the data is time-based. Segmented regression tries to circumvent cases where one cannot find one functional mapping. Instead, one tries to approximate disjoint intervals of the data with different models. Having disjoint intervals is really the only thing in common to Segmented regression and time-series segmentation.

In short: Time-series analysis is a classification of (time-based) based data, regression is fitting a model to the complete dataset. —PapaNappa (talk) 16:31, 8 December 2015 (UTC)Reply