Re: Stumped on crossover/mating...

Point taken. Still, this is going to keep me up a night. Can you
offer any insight in to how to solve it?

David vun Kannon wrote:
Hi Tanari,

I don't think crossword solving is a good first problem to attack when
learning GAs. Consider the following points:

- to solve the whole crossword will require the co-evolution of many
"species"
- the cardinality of the alphabet is very high. Most GAs work on a
binary alphabet, but you are working with a 26 letter alphabet
(assuming English words). This will require a large population to
ensure that all the values are available at every gene position.
- the representation allows many phenotypes (words) that are illegal.
- fitness value based on letter matches is going to be a coarse
grained measure for small words.

So, just because the evaluation function is easy to write doesn't make
this a good subject to learn GAs on. Instead of crosswords, you might
want to look into solving sudoku puzzles. But I would recommend trying
to write GAs for more traditional problems first, to familiarise
yourself with the general issues of representation and operator choice.

Cheers,
David vun Kannon

.