Complete.Org: Mailing Lists: Archives: freeciv-ai: December 2002:
[freeciv-ai] Re: RFC: Behaviour constants.
Home

[freeciv-ai] Re: RFC: Behaviour constants.

[Top] [All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index] [Thread Index]
To: Gregory Berkolaiko <Gregory.Berkolaiko@xxxxxxxxxxxx>
Cc: Freeciv AI development <freeciv-ai@xxxxxxxxxxx>
Subject: [freeciv-ai] Re: RFC: Behaviour constants.
From: banjo <banjo@xxxxxxxxxx>
Date: 05 Dec 2002 21:02:11 +1300

On Thu, 2002-12-05 at 11:27, Gregory Berkolaiko wrote:
> How can we improve AI without going too deep?
> 
> This is a legitimate question, and an answer to it is given by various 
> constants and factors and multipliers that the AI code is peppered with.  
> Examples are: what is considered to be a small nation (number of cities) 
> -- used to encourage settler building; what is the default danger for 
> capital; what is food/shield weighting etc.

This is similar to a the idea of Genetic AI's...

Have many optional weightings in the AI code, and define a genome as
being the concatinated list of valid options.

Then we can pitfight, mutate, breed and do all the standard GA tricks
for evolving fit competitors. Several classics may be distributed with
standard Freeciv others can be swapped, downloaded etc.  Or even defined
in Nations Rulesets...

  Shaka    d459v4539jwv045n5iko
  Neville  78c3b423bJ98f494igek

Modpacks can of course have their own species, more adapted to those
environments.

> A note on pitfights: when I did it, I used Per's script, but slightly 
> modified.  I would run a gmame with two players H and E, and each map 
> would be run twice, the second time with E in H's starting position and 
> vice versa.  Then if E would lose both games I knew my changes were making 
> it weaker; if E won both games, it was becoming stronger.

Now that AI can do Diplomacy, Pitfights should involve more than 2 AI's.

> I think this is the wrong answer. We can tune the AI a lot to a bunch
> of constants, but the moment we change the rules or use a modpack, 
> these constants will be way off. Better teach the AI to utilize the 
> rules better. Better attack code, better ferry code, better settler 
> code and better production code is much more of an improvement in the 
> long run than even a 50% improvement in pitfight stats due to tuning 
> of constants.

I agree totally, and yet their will still be many places in the code
where tradeoffs are to be made, and where it's a guess as to whether
greater agression is better than consolidation etc.

-- 
  \_   banjo 
  </'  banjo@xxxxxxxxxxxx
  /)   love songs * computer code * paranoia
 (/    Structure is strategy in slow motion!
 `



[Prev in Thread] Current Thread [Next in Thread]