Re: [glob2-devel] Re: Full Nicowar behaviour assessment

[Bradley Arsenault]

On 2/10/06, Bradley Arsenault <address@hidden> wrote:
> On 2/10/06, Stéphane Magnenat <address@hidden> wrote:
> > > Yes, i've been thinking about that too, but how? How could i
> > > guestimate avaialble resources? Maybe if i kept track of chunks of
> > > ressources, but then how do i decide which are mine and which are my
> > > enemies?
> >
> > Here is an idea, it's an approximation and I don't know if it works well in
> > practice. If you divide the maps in swaure of, let say 16x16. You consider a
> > square "belonging to you" if there is a building in it. You then compute the
> > density of food in all the square belonging to you. This is not very 
> > precise,
> > perhaps the algorithm should be more intelligent, by taking into account
> > ressources blobs connected to those squares or by taking into account every
> > ressource that is up to a certain distance to a building. You can compute a
> > mask of position to take into account and update the mask only on building
> > creation/destruction. Furthermore, updating this information is not very
> > frequent as globally ressources do not disappear in one tick.
> >
> > Have a nice day,
> >
> > Steph
> >
> >
> > _______________________________________________
> > glob2-devel mailing list
> > address@hidden
> > http://lists.nongnu.org/mailman/listinfo/glob2-devel
> >
>
> I actually just came up with not a bad idea. First, pregenerate a list
> of all chunks of ressources that are close enough to water to grow, so
> naturally ignored wheat that is inland and will dissapear quickly.
> Keep this list updated throughout the lifetime of the ai. Then, for
> each building, keep track of the nearest ressource that it is using.
> For inns and swarms, this would mean the nearest block of wheat, up to
> a certain distance. Then, when your deciding whether to "conserve" a
> particular section of wheat, you devide how much wheat there is in
> total by how many buildings are using it, and if its lower than a
> certain threshhold, begin conserving, primarily by sending messages to
> the modules cosntrolling those buildings to turn down usage, and by
> starting ressource farming on that section.
>
> The method is slow but accurrate. I'm big on slow algorithms, but
> sicne i'm on very fast hardware, its hard for me to determine if
> nicowar is going too slow.
>


Actually, I just did a profile, and even with 3 Nicowars battling it
out, Nicowar never took up more than 11.3 % of total runtime. I'm now
quite confused, for a while I have been thinking that it was Nicowar
causing the great amount of ressource consumption late in the game,
but perhaps I have been wrong. If someone else could also do a
verifying profile, it would be nice.