Jump to content

Talk:Iteratively reweighted least squares

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

Wrong objective function?

[edit]

It looks like the objective function stated at the very beginning of this article is incorrect. The objective function given there is just a weighted sum of squared errors, which is the objective function used at each weighted least squares step of IRLS, but not the class of objective functions addressed by the algorithm.

It is also misleading to write "It solves objective functions of the form...". It would be clearer and more accurate to instead say something like "It solves optimization problems of the form...". That is, IRLS does not solve the objective function. Rather, it solves the minimization problem with that objective function. I do not feel qualified at this point to actually edit the article, but perhaps someone who does feel qualified will see this comment. If nobody makes the edits after a while, I will decide my expertise is sufficient and try to make the edits myself.

Jasonrudy (talk) 21:13, 16 March 2014 (UTC)[reply]

I solved it for the moment by taking out some words and fusing two sentences. Problems still remain: The lead is too long, should be one or two paragraphs preferably devoid of (big) formulas, one should be able to read it out loud. There is scarcely a description on how the method works, and even less on why it works, on when one can expect what kind of convergence. Which should be essential for a method that has "iterative" already in its name.--LutzL (talk) 09:57, 17 March 2014 (UTC)[reply]
And feel free to edit, this is not an ideologically entrenched article. Just stay encyclopedic and on topic, that is, make generous use of links for topics that are not directly core parts of the method.--LutzL (talk) 10:02, 17 March 2014 (UTC)[reply]
Thanks. I think the wording is more correct now. However, I still think the first equation is incorrect. For example, the L^p norm linear regression example later in the article contradicts it. Unfortunately, I do not know the general class of problems that can be solved by IRLS. It's something I've been trying to find out, and one of the reasons I was looking at the IRLS article in the first place. If I do find out then I will fix the article. Jasonrudy (talk) 21:03, 21 March 2014 (UTC)[reply]


The last link does not work — Preceding unsigned comment added by 88.217.12.141 (talk) 07:38, 1 May 2014 (UTC)[reply]