Re: a doubt

[Subbarao Kambhampati <rao@asu.edu>]

At 11:33 PM 4/3/2003 -0700, you wrote:

> Dr Rao,
> Isn't h ind the same as h sum (without any adjustments for interactions )
> then ? 


yes


> another doubt , can u
> elaborate on relevence based pruning of no goods i.e what 
> is the basis of pruning a no good when it differs in more than k places
> from a 
> current assignment ?


Since I have already given the midterm, answering questions about the
topics that are
specifically on the midterm is sort of inappropriate (would be too much
of an embedded exam taking
for me). You can however read the relevant paper to figure it out (the
journal version of intelligent
backtracking paper that I referred you guys to has specific discussion of
it, as well as the 
original paper by Bayardo and Schrag)

rao




> Quoting Subbarao
> Kambhampati <rao@asu.edu>:
>
> > Suppose the cost of each literal li is ci
> > 
> > hind assumes that the cost of a set of literals {l1...lk} is sum of
> the
> > 
> > costs c1..ck
> > This makes sense only if the subgoals are completely independent
> (i.e.,
> > 
> > achieving
> > one has no positive or negative effect on the achievement of
> the
> > others).
> > If the subgoals are positively interacting, the actual cost will be
> less
> > 
> > than what hind computes
> > (so hind is not admissible). IF they are negatively interacting,
> the
> > cost 
> > computed by hind can be
> > grossly lower than the real value (e.g. you may say Holding(A) can
> be 
> > achieved in one step
> > holding*B) can be achieved in one step. The two together cannot
> be
> > achieved 
> > with any finite
> > length plan (so their cost is infinity--although hind says it is
> 2)
> > 
> > hlevel looks at the first level of the PG where the goals come
> in
> > together. 
> > This take into account
> > both positve and negative interactions into account. IN the holding
> case
> > 
> > above, hlevel will come out as infinity
> > (for normal pg with mutex relations).
> > 
> > Rao
> > 
> > 
> > At 10:08 PM 3/31/2003 -0700, you wrote:
> > 
> > >Dr Rao,
> > >What is the difference between hind and hlev heuristic ? Why do
> u say
> > that 
> > >hind
> > >is inadmissible  ?
> > >
> > >Sincerely
> > >Preetha
> > >
> > >
> > >
> > >
> > >Quoting Subbarao Kambhampati <rao@asu.edu>:
> > >
> > > >
> > > >
> > 
> >