Sunday, September 30, 2007

there is no typo/erro in the pattern database question..

Gavin's comments notwithstanding, there are no typos in the pattern database
question. You should do it the way it is given

(by the way, it is fine for h*(n) to be \infinity. You can still talk about admissiblity and
relative informedness of h1(n) and h2(n) ).


1 comment:

Gavin Lewis said...

I was thinking about ways to define h* in two problematic cases.

1 - no goal states exist.
2 - no reachable goal states exist (subset of 1).

The problem seems to be that h* is defined in terms of lowest cost, and that cost is somewhat problematic in the 'no reachable goal states' case (simply defined as infinity there?) and is more problematic when there are no goal states.

When there are no goal states, does it make sense to say that h* has no value, or alternately that h* has any value? (Let's say we pick 10 for h*. There exist no goal states with higher cost, because there exist no goal states.)

If we take a different approach and require h* to correspond to an actual path or set of paths with the same cost (need not be explicitly known), it has no value for either of the two cases.

What I'm guessing is that we're simply defining the cost of an unreachable goal to be positive infinity, and we are excluding "no goal states exist period" from the domain of h*. Does this sound right? [It feels like the best balance of pragmatism and consistency.]