A few scattered comments to our textbook, chapter 7

Section 7.3: Notice that algorithm on p. 222 and diagram p. 223, refers to notions such as "crossover" and" mutation" which are defined later; similarly "crossover probability", p. 224.
P. 227. A headline seems to be missing before "In this example...". Assume headline "Now to something different" ;-)
P. 228. The "conversion" from binary to decimal is confusing; no programmer would ever do like this!

Section 7.4: Should be ignored as arguments are unclear and mathematical arguments based on tacit and problematic assumptions. (Don't read)

Section 7.5: Example is rather construed. For the example to give sense, one must assume that "year" consists of four contiguous "intervals" (rather artificial, but never mind). Thus units 1 and 2 of the example are expected to be under maintenance 50% of the time, but never mind.
"Pool of genes" at p. 237 for units 1 and 2 seems to lack genes 1001.
Performance graphs 7.11 and 7.12 are strange. Fitness is defined to be 0 or above in the text, but these graphs indicate values down to minus 10. Furthermore, a comparison of 7.11 (a) and 7.11 (b) indicates a bug. The first half of (b) should be identical to (a) (qua a different scale on x-axis), but fitness is higher in (b) that in (a) which should not be that case.

Section 7.6 is difficult to grasp as terminology changes. What 7.6 calls "population" is the same as "chromosome" in earlier sections; similarly "solution" is used for what used to be called "fitness". Compared with the previous, section 7.6 works with a population of one chromosome.

Section 7.7 on genetic programming.
Please ignore the sentence "The aim of genetic programming is ...".
The algorithm explained in the text (p. 250) does not employ the fitness function in any clever way. It simply generates random offsprings without any consideration about fitness (and the solution found is the one which, in this random history, has had the highest fitness).
When the (supposedly) same algorithm is explained in a diagram, p. 252, fitness is applied to copy the most fitted so far into the next generation, but fitness is still not taken into account for selecting parents.
It is not clear whether this is a mistake or there should be some good reason for it which is not explained.


Last modification 9-oct-2007, Henning Christiansen