[Freeciv-Dev] Re: [RFC] Path finding interface #9

[Raimar Falke <rf13@xxxxxxxxxxxxxxxxx>]

On Wed, May 29, 2002 at 11:49:17AM +0100, Gregory Berkolaiko wrote:
> On Wed, 29 May 2002, Raimar Falke wrote:
> 
> > On Tue, May 28, 2002 at 08:07:17PM +0100, Gregory Berkolaiko wrote:
> > > Raimar,
> > > 
> > > 
> > > On Tue, 28 May 2002, Gregory Berkolaiko wrote:
> > > 
> > > > On Tue, 28 May 2002, Raimar Falke wrote:
> > > > > 
> > > > > Is it true:
> > > > >  - if we remove the is_position_safe and let the trireme issue out
> > > > >  that the best path is independent of the choose policy (TC_MINIMUM,
> > > > >  TC_AVERAGE, TC_MAXIMUM)?
> > > > 
> > > > no: you have 6 moves/turn, choose between these 3-segment paths:
> > > > (a) 5-6-5
> > > > (b) 6-6-2
> > > > (c) 5-2-6
> > > > 
> > > > MINIMUM will go for (a)
> > > > MAXIMUM will go for (b)
> > > > AVERAGE will go for (c)
> > > 
> > > Personally I see no sense in these 
> > >   turn_cost_factor, move_cost_factor.
> > > Why would you ever need such clumsy and complicated function as
> > >   turn_cost_factor * turns_needed 
> > >   + move_cost_factor * (move_rate - moves_left)
> > >   + sum(basic_move_cost * PF_BMC_FACTOR + extra_cost) ???
> > > 
> > > In my opinion 
> > >   move_points_spent + sum(extra_cost)
> > > is enough, if move_points_spent is calculated according to one of the 
> > > 3 models: MIN, MAX and AVE.
> > > 
> > > If you disagree, please give examples.
> > 
> > Suppose you have two paths. The one is short and dangerous (higher
> > extra_cost) and the other is long but not so dangerous (lower
> > extra_cost). You than need a method to trade danger against extra time
> > for the path. As we have seen BMC isn't enough to capture the length
> > of a path.
> 
> This is perfectly catered for by the formula
>       move_points_spent + DANGER_WEIGHTING * sum(extra_cost)
> 
> However, as extra_cost function is user-supplied, the user can (and 
> should) incorporate the DANGER_WEIGHTING into the extra cost.  Hence the 
> final quantity to minimize,
>       move_points_spent + sum(extra_cost)
> 

Ack.

> Just a reminder, move_points_spent is not merely a sum of BMCs but the 
> estimation of the length of the path, in move-points.

Ok. And can I say that I want to trade saved turn for x "danger
points"? The difference where I arrive with n turns and 5 points left
is different to arriving in n-1 turns and 1 point left. It is more
different than the two move points.

        Raimar

-- 
 email: rf13@xxxxxxxxxxxxxxxxx
  reality.sys corrupt. Reboot Universe? (y,n,q)