Talk:Harmony search


 * "1. HS does not require complex calculus, thus it is free from divergence."

Either my understanding of "complex calculus" and "divergence" is completely wrong, or this is nonsense. 217.226.83.152 (talk) 14:05, 1 January 2009 (UTC)

Needs verification
This page has:
 * No description of the algorithm
 * At most two or three major contributors


 * Yes, really. The link ("Harmony Search Algorithm" doesn't really give any more explanations (except maybe in Korean), it seems to promise a lot, but it's on an analogy-level and stops ther (harmony, tuning, pitching, ensemble...; analogies are nice, but in the end analogies are all they are).  In the Excel-link on the mentioned page, I see a few (mainly uncommented) lines of VBA-code that tries the Rosenbrock test, with sometimes (randomly) unimpressing results.
 * There must be more to it?--80.202.212.89 14:11, 23 June 2007 (UTC)


 * I agree, I looked at a few papers and they have a lot of self-similarity and self-citation. The convergence on the rosenbrock test is also underwhelming (it took 100000 iterations to get within 1e-5 of the optimum - I think Nelder-Mead method is a few orders of magnitude faster than this (the animation on that page has 45 frames). --Jaded-view (talk) 21:43, 28 April 2008 (UTC)


 * Agreed with all of you -- this looks like pseudoscience to me. "divergence free" ?? The author either doesn't know calculus, or needs an improved translation (it's clear that the main article was written by a non-native speaker).
 * Also, the algorithm description doesn't make sense. The letter $$i$$ is used as a superscript and a subscript, so it's not clear if it's referring to an index or an iterate or both. Lavaka (talk) 03:08, 15 September 2009 (UTC)

Genetic algorithm?
I've played around with harmony search a little. The algorithm is not well described in the publications I was able to source, but I inferred it from the Excel example. It seems to work OK on toy engineering problems.

Its closest cousin is the genetic algorithm. In fact it can be considered as a type of GA (depending on how tightly you define a GA). In any case, it's a form of stochastic optimisation, not gradient based.

A conventional GA works by crossing two parent genotypes to produce offspring. Thus each "gene" in the offspring's chromosome (usually a 0/1 bit) comes from one of the parents. In harmony search each gene is selected from the population at large, so a "child" might have numerous "parents" contributing to its genotype. At least that's the way I implemented it.

As an amateur musician and composer myself I found the "harmony" analogy unhelpful.

--88.109.101.42 (talk) 17:39, 25 November 2009 (UTC)

Criticism
A critical review can be found in: A Rigorous Analysis of the Harmony Search Algorithm: How the Research Community can be Misled by a “Novel” Methodology by Dennis Weyland (Istituto Dalle Molle di Studi sull’Intelligenza Artificiale (IDSIA), Switzerland)  82.79.21.191 (talk) 16:09, 9 February 2011 (UTC)


 * I have removed the boilerplate criticism section that appeared in the lede of this article (and in many others). I have replaced it with a section that covers only the criticism specific to this algorithm, and refers the reader to the more general criticism of nature-inspired metaheuristics at List of metaphor-based metaheuristics. Since I have made similar removals or replacements on other pages, if anyone has an issue with this one, it's best to start an omnibus discussion at Talk:List of metaphor-based metaheuristics. —Psychonaut (talk) 19:13, 31 August 2016 (UTC)


 * The "Criticism" section of the article is presently tendentious in tone, and in no way neutral. This particular article needs improvement, whatever the merits of the "omnibus discussion" suggested by User:Psychonaut – which I'll also try to look into.  yoyo (talk) 17:57, 16 December 2016 (UTC)

Deletion
This algorithm is almost a textbook case for the worst practices in metaphor-inspired algorithms. So that sounds like a perfect argument for deleting the page, right? I strongly disagree. To paraphrase a common argument for the (US) first amendment, the best remedy for foolish speech is more speech. Leaving this article here, with clear citations to the work of Weyland and others demolishing the value of this algorithm, is the best way to alert future readers to the pitfalls of the algorithm - and it's far more likely to provide early warning than a literature search, which will still (2018) turn up so many positive citations that new readers (and researchers) are likely to be misled. — Preceding unsigned comment added by Urilarim (talk • contribs) 07:12, 1 December 2018 (UTC)