Re: Tree search
- From: berndlosert@xxxxxxxxxxxx
- Date: 31 Aug 2006 09:23:49 -0700
Since your procedure traverses the tree from the root to some leaf
node, wouldn't the average amount of steps required for a solution be
the sum of the steps required to reach every leaf node, this divided by
the number of leaf nodes?
--
Bernd
kingpin@xxxxxxxxxxx wrote:
Hi to all,
you have a problem. Its search speace is modeled with a tree with
finite height. You have an evaluation function (heuristic) that says,
at every level i of the tree, if a node is better than another. The
heuristic is not perfect in the sense that it has some error. The
solutions is a path from the root to a node on the leaves that has some
property. Obviously, you don't heave enough information to decide
anything else: you must search.
I want to use A* (or best-first search) and study the aveage number of
nodes expandend before a solution is found.
How should i proceed? How should i specify the error the heuristic make
in order to simplify the analysis?
For example: the heuristic h is strictly monotonic, the tree is binary
and the problem is a minimization one.
I start to search with a b-f search. If, at level i, i make the wrong
decisions, i must search the whole sub-tree rooted at the son selected
before i can use the right node:
[i] level i
| \
| \
[1] [2] sons of level i
Since h([1]) < h([2]), than for every node n in the subtree rooted at
[1], h(n) < h([2]).
So, the "delay" is exponential....
I'am interested in the case that h is not strictly monotonic....
Any suggestion? Any paper? Any book?
Thank you,
Lucas
.
- References:
- Tree search
- From: kingpin
- Tree search
- Prev by Date: Tree search
- Previous by thread: Tree search
- Index(es):
Relevant Pages
|