Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Programmers: ETC

Author: Robert Hyatt

Date: 21:01:01 02/13/03

Go up one level in this thread


On February 13, 2003 at 16:30:04, Antonio Dieguez wrote:

>On February 13, 2003 at 15:23:15, Peter McKenzie wrote:
>
>>Just wondering how many programs are using the Enhanced Tranposition Cutoffs
>>algorithm, or if people have tried it and had trouble with it.  I think I'll
>>have a go at implementing it shortly.
>>
>>For those interested, here is the basic idea:
>>
>>When searching a position, the first thing done is a search of the hash table.
>>Assuming this doesn't result in a fail high, we then try ETC.
>>
>>For each possible move, do a make move and check the hash table.  If the hash
>>table indicates that the side to move would fail low (it must have sufficient
>>draft of course) then you can fail high in the original position.
>>
>>I guess the tricky thing is making sure that alpha and beta are handled
>>correctly.
>>
>>There are possible tricks to make it work better:
>>- only do ETC when remaining depth is reasonably high
>>- have a special fast makemove which updates the hash key only
>>- even if ETC doesn't provide a cutoff, perhaps it can help move ordering
>
>I tried etc a long time ago... (since am young, a long time isn't much)
>I didn't try hard though so I will try harder now, thanks for mention it.
>
>there is another tricky thing I just realized after seeing a few searches
>changing too much there and there. In amyan the depth saved in the hashtable is
>not always the original depth. As I do the extensions first, and forget the old
>depth remaining when probing and saving in the hashtable. I never saw a drawback
>with that but now I see it: etc. I will change that and is easy to do it, but I
>wonder if the other way has another drawback, after all the same depth will be
>saved even if the node extended or not in this ocassion, and it seems it affects
>the extensions even more but perhaps am wrong or it doesn't matter much.
>
>be well.
\


All that is critical is that you store the same depth you probe with. If you
do some extensions before storing, you had better do those same extensions
before probing.  IE if you don't, you are going to introduce some huge artifacts
that will break things.

Either you will be cutting off when you should not, or not cutting off when
you should.  The former will cause the search to become very unstable.  The
latter will make it slow.



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.