On 10/21/07, Justin Giboney <[EMAIL PROTECTED]> wrote:
> Thank you Victor for your kind, and professional response. Sorry, if I
> offended anyone. I didn't want to necessarily incriminate anyone. I just
> wanted to kind of find out if someone had any theories on doing this
> test, or thoughts behind it. If they did it for research than I am all
> for it.
>

justin,

this test will always tend toward brown. as the number of generations
approaches infinity, the resultant colors approach brown... and i
don't think any of the voters actually prefers brown.

i'm kinda tired, so i don't know how well i can explain this, but i'll
give it a shot: i would guess that the problem is with the assumptions
for the test. say, for example, that all participant preferences are
equally distributed, but all participants prefer bright colored links:
participant 1 prefers bright red, participant 2 prefers prefers bright
green, and participant 3 prefers bright blue. by genetically combining
these results, we get a slightly browner (and dimmer) generation. if
every set of participants has the same preferences, the resultant
colors will become dimmer and browner with each successive generation.
but remember that the participants all like bright colors.
unfortunately, there's no way to learn "brighter" as any combination
results in "dimmer".

maybe instead of r/g/b genetic combinations, try using independent
genetic pools for hue, saturation, and brightness. in each test,
present the user with either a set containing a variety of hues, or a
set with a variety of brightnesses (each with the same color), or a
set with a variety of saturation levels, each with the same brightness
and hue.

justin
-- 
http://justinhileman.com

_______________________________________________

UPHPU mailing list
[email protected]
http://uphpu.org/mailman/listinfo/uphpu
IRC: #uphpu on irc.freenode.net

Reply via email to