Talk:Genetic algorithms

Reviewer A

This is a concise and very useful characterization of "genetic algorithms." i'll hope others are fleshing out other parts of "evolutionary computation."

my primary object is to the last para of the "Crossover" section. It's use of "schema theorem SHOWS...", "... EFFECTIVELY PROCESSES ..." , especially directly invoking Fisher's fundamental theorem makes the reader think both are equally well-established. i don't think that's the case, as yet. There has been an extended debate over this issue that should be acknowledged. Mitchell's treatment is a good starting reference. Grefenstette, Whitley (GENITOR), Vose93, L. Altenberg...

or, narrow the claims: "... alleles present in the INITIAL/EARLY populations ..."; "... fitness spreadk rapidly through the population UNTIL ...?"

and/or, relate to other machine learning methods for "effectively processing" statistical quantities; graphical models (aka Bayesian networks) would make for an excellent contrast.

the other substantive change i'd recommend is including a sentence regarding alternative choices for the NUMBER of Xover points used, vs. the one-point assumed here. i favor 2-point, because of its utility with CIRCULAR chromosomes.

In general, it is well written and needs little editting. "... if the parents’ specific chromosomes are never passed on?" ends on a preposition, but perhaps this is unavoidable.

Reviewer B

This is a nice concise summary of the basics of genetic algorithms. In general it is quite appropriate.

• This is a bit of nitpicking for what is a fairly small error, but my concern is a mischaracterization of crossover in a biological context. This sentence is incorrect:

"Crossover is the central reason that mammals produce offspring exhibiting a mix of their parents’ 
characteristics – consider the human population for a vivid, familiar example – and crossover makes
possible the artificial cross-breeding of selected animals and plants to produce superior varieties."

Crossover has no relation to blending of parental characteristics in diploid organisms, those having two copies of each chromosome (which includes nearly all animals), and crossover as generally thought of occurs almost exclusively in diploids. Crossover happens at a particular stage during meiosis, when an individual divides up its genetic material to be packaged into haploid gametes (having a single copy of each chromosome, i.e. sperm and eggs). Meiosis is a within-individual process and has no relation to a specific mating or to creation of a new individual from the fusing of parental gametes. Its effect in a particular individual is to blend the chromosomes contributed by that individual's parents before passing them on to that individual's offspring. An individual always receives 1/2 of their genetic material (one copy of each chromosome) from each of their parents, and this is the cause of blending of characteristics. Without crossover, the 1/2 that comes from each parent would be identical (baring mutation) to one of the two chromosomes passed on to that parent from its own parents.

Diploid crossover is indeed critical to biological evolution because of its ability to combine the information on separate chromosomes (~solutions), and as such it is a decent analogy for the effects of GA crossover. However, typical biological crossover is not a good analogy for GA crossover in some of the mechanistic details, in particular in when it occurs (not during mating), as well as in the fact that it occurs in diploids. Bacteria (generally haploid) can in unusual circumstances undergo something more akin to GA recombination, but in general there is no biological case that I am aware of in which two "parents" having a single chromosome each come together to create children with a single chromosome each by recombining the parental copies.

The main change required by this issue of the timing of crossover is to remove the first 2/3 of the above sentence, leaving that after the second hyphen. Simply removing "mating" from the following sentence should also suffice.

"This simple version of crossover is a good approximation to what actually takes places in mating organisms."

The rest of the discussion, including the effects of linkage, is all accurate.

• One other issue, the picture of crossover appears to be of the cut and splice type, not one-point as referred to in the text. The picture shows different cut points in each chromosome, and a change in chromosome length, which isn't possible in one-point as I understand it. One-point

is a good approximation of biological crossover, but cut and splice is not.