Computer Chess Club Archives


Search

Terms

Messages

Subject: Fractional Iterative Deepening search

Author: Gian-Carlo Pascutto

Date: 12:50:43 06/18/00



Hi all

In the discussion of the 'Scalable Search Test' thread with
Ed Schroeder I mentioned that MTD(n,f) has the nice property
of making a fail-high pretty constant over time. I.e. the
search does not blow up as it does in a normal PVS searcher.

Unfortunately it seems that this does not help when moving
up a ply...it even seems that the results of the MTD'ers
are quite terrible.

The following though occured to me, if MTD allows you to take
small steps in the score plane, what about using fractional
ply increments to take smaller steps in the depth plane?

Many of the best programs have now switched to fractional extensions.
Thus, fractional search depth must make sense.

Iterative deepening is one of the most important improvements to AB
search. Thus, it makes sense too.

Still, the programs use whole ply's in their iterative deepening
search. Why? It would make perfect sense to step in smaller increments
too. I feel this can even give improvements in tactical situations,
where the fractional extensions are triggered.

I'm interested if someone has ever done or tested this before. Did it
work? What were the results?

If you happen to have a program which uses fractional extensions, please
try it, and let us know how it works out.

--
GCP



This page took 0 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.