[freeciv-ai] Re: Hi, I'm back. aiclient

[Raimar Falke <i-freeciv-lists@xxxxxxxxxxxxx>]

On Mon, Jun 30, 2003 at 11:37:31AM +0200, Manuel Gutierrez Algaba wrote:
> El Lun 30 Jun 2003 10:31, escribió:
> 
> > int base_get_attack_power(Unit_Type_id type, bool veteran, int moves_left)
> > {
> >   int power;
> >
> >   power = get_unit_type(type)->attack_strength * POWER_FACTOR;
> >   if (veteran) {
> >     /* Veterans get +50% bonus. */
> >     power = (power * 3) / 2;
> >   }
> >
> >   if (!unit_type_flag(type, F_IGTIRED) && moves_left < SINGLE_MOVE) { ; I 
> don't understand this rule very much, I'd have to read more about 
> unit_type_...

unit_type_flag(type, F_IGTIRED) corresponds 1:1 to the 
; "IgTired"     = ignore tired penalty when attacking
flag in units.ruleset.

 > >     power = (power * moves_left) / SINGLE_MOVE;
> >   }
> >   return power;
> > }
> >
> Well, If I remember correctly Clips, would be:
> 
> Using Clips, an expert system language:

[snip code which looks like lisp ]

> Apparently is even more complex.

It is harder to understand. This may not be because C is better/worse
than clips but because I (and almost all other developers) have
learned how to read C. So C has an advantage.

> But let's see it integrated in the "whole":
> 
> - When you call the C function you have to store the "power" value
> and take a decision *promptly" somewhere. Somehow it provokes that
> the decission must be taken soon, and in an "if" branch up in the
> stack of calls.
> - C is a model that has only the info to be used at the moment, and
> to use some info (attack power) you have to request for it.

There is confusion here. IMHO we have to seperate "rules" and
"decisions". "Rules" are inherent freeciv rules like "Veterans get
+50% attack bonus." They don't change much. They should be keep in
only one form (in the C source code in common/).

There are "decisions" like "is it a good idea to irritate this
tile?". Answering these questions has something to do with the term
"AI". C may be not be the best language to code this.

> Now Clips mode:
> - Whenever anything changes in the game (a unit becomes veteran) ,
> all the related facts about it , such base-get-attack-power,
> defense... become updated at the moment. That provides a pool of
> asserted facts ready to be included in some rules of high
> order. Like the fact defense-power-of-city, attack-power-of-fleet...
> I mean that in Clips model , you have all the info concurrently
> ready to be used, you don't have to call any function to work out
> values and then produce a single "isolated" value, that most likely
> will be consumed within a "chain" of decision.

I don't know the correct terms but I would call this "event driven":
an event can update state and generate so more events. It is an
bottom-up strategy. The other one is what I would call "demand
driven": to answer a question you divide the problem and solve each
sub problem. This is a top-down strategy. This is orthogonal of the
clisp vs c and the "rules" vs "decisions" discussion. While I agree
that C is better suited to the top-down strategy.

> In Clips you'd have "tree" of decisions and "trees" of reactions. A single 
> change will recreate a tree of facts.
> 
> Needless to say, this model is richer and thought-provoking. The
> logic may seem the same , the difference is in the data: more and
> more interrelated and usable.

The question is how can this be transform into a real advantage: the
AI is easier to program, the AI is faster, the AI is stronger. While
it is another model I don't see that the advantages are an
implication.

> The difference (C-Clips) doesn't come in isolated rules, but in the
> behaviour of sets of rules. In C they're line-chained, call-by-call,
> in Clips are "explosion"-like , a single change trigger changes in
> many variables (decissions to be taken). In C you focus on the
> "function" the call, in Clips you focus on the state, the fact. C is
> imperative, Clips "logic" or better "stative". With the same effort,
> because the logic for producing a value/fact is the same in C or
> Clips (you somehow must generate the value with if's or defrules),
> in Clips you keep a "trace" of facts and a set of facts, that in C
> would had remained as local variables or returned values, hard to
> relate, once the "initial call" has ended.
> 
> Prolog would be pretty much the same, and better, since Prolog can
> "fill" the gaps and "search for" a solution, not simply fact-making.

I would like to see a true aiclient (one that does only the
"decisions" from above) in this model. Even a simple one (build
settlers, build cities, explore the map). 

        Raimar

-- 
 email: rf13@xxxxxxxxxxxxxxxxx
 "Many of my assistants were fans of Tolkien, who wrote 'Lord of the Rings'
  and a number of other children's stories for adults.  The first character
  alphabet that was programmed for my plotter was Elvish rather than Latin."
    -- from SAIs "life as a computer for a quarter of a century"