NetNews Usenet Archive 1993 #1/NN_1993_1.iso/spool/comp/ai/genetic/29

Text File | 1993-01-11 | 1.7 KB | 39 lines

Newsgroups: comp.ai.genetic
Path: sparky!uunet!usc!sol.ctr.columbia.edu!eff!news.byu.edu!ux1!fcom.cc.utah.edu!mail.physics.utah.edu!freier
From: [e-mail] (rodney james freier)
Subject: Re: Simple problems to solve genetically
Message-ID: <[e-mail]>
Sender: [e-mail]
Organization: University of Utah - Physics Department
Date: Mon, 11 Jan 93 16:20:55 GMT
Lines: 28

  Hello.

   As some of you have pointed out, my lengthy analysis of the blackjack
problem overlooked some obvious optimizations.  For example, blackjack does
not care one whit about the suit of the cards, only their value.  So, instead
of needing to keep track of the 52 cards, it only has to keep track of 10
different values.  Then we have:

     Hand's value         9
      dealer's upcard     10
      my first card       10
      my second card      10

 for a total of 9000 bits for the strings.  This is actually workable.  For
five cards, the total approaches 10k.  This is pushing it.
    (My insistence on 52 cards is what is technically known as a Brainfart(tm).)

    Some of you have mentioned much simpler counting methods... both simpler
to code and simpler for a human to use in play.

     I agree that all this work seems to have taken some of the fun out of
making a computer learn.  But I don't believe I have done the significant
learning.   Actually, what I'm doing requires and results in
NO knowledge of when to hit and when to stay.  All I am doing is trying to
make sure that my string pattern can represent ANY conceivable strategy, and
the mechanics of selection then search this space to find effective strategies.

Rod