Re: Using partial plys

[David Fotland <fotland@xxxxxxxxxxxxxxx>]

There is little similarity to B*.  My search is trying
to prove something quite different.  B* is trying to
prove separation between the best move and second best
move using optimistic and pessimistic valuse, and probabilities.

I am reading life and death, so I am trying to find a move that
works.  The result at leaf nodes is either success or failure, so
there is no evaluation function value being backed up.  When I
find a move that works with high confidence I can stop.  My 
probabilities are confidence values in the result.  When I expand a
node, I expand a single unbranching line of play to a leaf that
I can reliably evaluate as success or failure.  B* expands one
node at a time, and uses small probe searches to get values
for the bounds.

David

At 10:12 AM 3/23/99 +0100, Tristan Cazenave wrote:
>David Fotland wrote:
>
>>  What information is stored for each search node
>> >and how is this information used to update ancestor nodes and direct the
>> >search.
>>
>> The result (win, win-likely, ko, unknown, lose-likely, lose), the
>> confidence (0 to 100), the confidence of the most refuting node below
>> this one, has this node been evaluated.  Confidence values are passed
>> toward the root, using the usual probability formulas.
>>
>
>Is there any similarity with B* Probability Based Search?
>It is a best first algorithm that also use probabilities.
>Put another way: what are the differences ?
>
>--
>Tristan Cazenave
>Universite Paris 8, Laboratoire d'Intelligence Artificielle
>2, rue de la Liberte 93526 Saint-Denis Cedex - France
>http://www.ai.univ-paris8.fr/~cazenave/
>
>
>