Here's an alternative (diploid) approach for you genetic algorithm:
ord=: a. i.]
chr=: a.{~]
fitness=: [: +/"1 [: | -"1
evol=: (2 :0)
:
'epsilon psize growth'=.m
goal=. n i.x
genepool=. psize$,.&.|:n i.y
glen=. {:$genepool
mutations=.(0#~3*glen*#n),(1#~#n),(_1#~#n),i.#n
grow=. growth*psize
0 goal track n genepool
while.1 do.
goal track n genepool
children=.(?grow$1+glen) (({. {.), (}. {:))"_1 (?(grow,2)$psize){genepool
children=.genepool,(#n)|children+mutations{~?($children)$#mutations
cost=.goal fitness children
ok=.epsilon >: cost
if. 1 e.ok do.
1 goal track n children
n{~ ~. (ok#children) /: ok#cost return.
end.
genepool=. psize {. children /: cost
end.
)
track=:2 :0
gCount=:gCount+1
if.1e4<gCount do. 1 m track n y throw.end. NB. safety valve
:
if.0=x do.gCount=:0 end.
min=.</\cost=<./cost=.m fitness y
1!:2&2 'gen ',(":gCount),'(',(":min#cost),') ',,n{~min#y
)
Example use:
(] 0 2 30 evol (chr 32+i.95) [:chr 32+[:?95$~#) 'A perfect creature!'
gen 0(549) Ya;ev}X^A)bG:zR)e{N
gen 60(0) A perfect creature!
A perfect creature!
Note that this runs somewhat faster if the references to track are
commented out.
Note that this returns a list of possible answers. (Of course, this list will
be length 1 when an exact match is returned.)
FYI,
--
Raul
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
