Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Getting a PV using MTD(f)

Author: Tijs van Dam

Date: 14:32:31 02/08/00

Go up one level in this thread


On February 08, 2000 at 12:40:02, Andrew Williams wrote:

>On February 08, 2000 at 12:21:18, Tijs van Dam wrote:
>
>>On February 08, 2000 at 11:01:00, Andrew Williams wrote:
>>
>>>On February 08, 2000 at 08:54:22, Tijs van Dam wrote:
>>>
>>>>I have been reading about the MTD(f) algorithm that is also used by Andrew
>>>>Williams in PostModernist. I want to try it, but it has some things different
>>>>than a "traditional" search. I am wondering about the following: how do you get
>>>>a PV, or even a move to ponder?
>>>>
>>>>The best move is obvious: it is the last move that failed high at root. But when
>>>>searching that move, all moves in the child position must have failed low. So
>>>>which one of them is the best? Andrew, how do you do this?
>>>>
>>>>My first thought was doing an MTD search with shallow depth from the position
>>>>that follows the root move. Or maybe a PVS search?
>>>>
>>>>Who on CCC has experience using MTD?
>>>>
>>>>Greets,
>>>>Tijs van Dam
>>>
>>>I use my hash-table to keep track of the PV. If I've found a move at
>>>a position whose score is > beta, that move ALWAYS goes into the hash-
>>>table. I then rebuild the PV by stepping through these moves. It works,
>>>but often you find the end of the PV being truncated.
>>>
>>>BTW in PostModernist I hash moves in the qsearch as well.
>>>
>>>Andrew
>>
>>Thanx, I'll try that. I've just converted my program to a first raw version, and
>>the results are so far so good. I think the hashtable storing needs some tuning.
>>Is hashing faster than researching the q-search for you?
>>
>>Tijs
>
>Yes, it seems so. It's so easy to take it out and put it back in that I test
>this quite often, and it's always been better with hashing in the qsearch.
>Did you have a look at the web page I posted in the "Green List" thread? In
>there it describes what you need to do to make your hash-table work with
>MTD.
>
>Good luck and please keep us posted.
>
>
>Andrew

Yes, in fact that's what gave me the whole idea of using MTD. I'm experimenting
a little, but I still use about three times as much nodes for calculating a
position with fixed depth as I do with conventional alpha-beta. I will work on
it this weekend and post the results.

Tijs



This page took 0.05 seconds to execute

Last modified: Thu, 07 Jul 11 08:48:38 -0700

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