[freeciv-ai] Re: (PR#4089) Estimated incite gain is too high.

["Gregory Berkolaiko" <Gregory.Berkolaiko@xxxxxxxxxxxx>]

On Wed, 30 Apr 2003, Per I. Mathisen wrote:

> On Wed, 30 Apr 2003, Gregory Berkolaiko wrote:
> > > I'll look at pf again, but the incite comment is dead on. Will you commit
> > > this?
> >
> > Since we are at it, I thought about other ways of estimating gain. Here
> > are my ideas.
> >
> > 1. Cost to take city by force is (a) hard to estimate and (b) misses the
> > point: what if we don't _want_ the city?
> 
> Taking the opponents' cities is one way to win the game. The most common
> way, too.
> 
> > 2. We should instead ask "will we get our money back in reasonable time?".
> 
> This ignores the fact that conquering cities isn't an investment, as much
> as a military victory over an enemy we want to vanquish.

Okay, you convinced me.

> > 4. To stop before we go too deep, use
> >
> > def_upkeep
> > = unit_types[best_role_unit(L_FIRSTBUILD)]->sheild_upkeep;
> >
> > benefit = food_surplus * FOOD_WEIGHTING
> >         + (shield_prod - 2*def_upkeep) * SHIELD_WEIGHTING
> >         + (luxury_total + tax_total + science_total) * TRADE_WEIGHTING;
> >
> > benefit *= planning_horizon;
> >
> > benefit -= incite_cost * TRADE_WEIGHTING;
> 
> I think this mostly misses the point of inciting cities. This is only if
> we incite cities of players that we are not at war with, and then doing
> this _will_ bring us into war pretty quick.
> 
> We want, first and foremost, to incite cities that are hard to take with
> our normal units, cities that support a lot of enemy units, and cities
> with important wonders.

But the current calculation is not doing that either!  We should then look 
at 
(*) the number of defenders
(*) defensive structures
(*) wonders
(*) production

The example I gave in the original email was that a size 2 city with no 
buildings or defenders was valued at 4182, which essentially means that AI 
is prepared to pay
        4182 / 12 = 348 gold
for it (inclusive of diplomat travel expenses).  I think it's overvalued 
(not by a large factor though).

> Will you commit my patch in the meantime?

Okay, okay...

G.