Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: History heuristic

Author: Tord Romstad

Date: 03:04:09 03/19/04

Go up one level in this thread


On March 19, 2004 at 05:55:17, Sune Fischer wrote:

>On March 19, 2004 at 05:43:07, Tord Romstad wrote:
>
>>I am not sure I understand the logic behind the last rule above.  Is there
>>any reason to believe that safe pawn pushes are better (in general) than
>>most other moves?
>
>I think maybe he means passed pawn pushes?

Yes, that makes more sense.

>I think once the obvious "try a quick cutoff move" has been searched, it would
>make sense to order the moves that gets extended to also be searched first. By
>'definition' they are interesting moves.

Perhaps.  But they also tend to lead to bigger subtrees.  If there are
several moves which would fail high, I would prefer to search a move
which does not cause an extension.

>>>Also I'm trying to implement some
>>>attacks info -- "forks" e.t.c. Hint: expensive knowledge can be implemented when
>>>remaining depth >2*INCPLY or >3*INCPLY e.t.c.
>>
>>Yes.  It's strange that so few people seem to realize this.  Apparently,
>>almost everyone uses exactly the same move ordering techniques at all nodes,
>>regardless of the remaining depth.  It makes sense to use much more expensive
>>move ordering knoledge when the remaining depth is big.  If the expected
>>size of the subtree is millions of nodes, it is clearly a good idea to
>>spend a lot of effort to make sure the best moves are searched first.
>
>If you have a good scheme you can probably benefit from it all the way to the
>leaves, perhaps only at the last ply or two it will be too expensive.
>IID is one such example btw.

You're right, IID is the most obvious example.

>>>:) It will be. But I'm waiting for ST that will be significantly stronger than
>>>Ruffian 1.05.
>>
>>Please don't wait so long! :-)
>
>Huh?
>
>SmarThink doesn't appear to be very far from Ruffian strength already.

Perhaps not, but going from not very far behind Ruffian to significantly
stronger than Ruffian is still a big jump.  :-)

>http://wbec-ridderkerk.nl/html/his6thedition.html
>
>>I once asked you whether there was any chance that you would port your
>>engine to Linux and/or Mac OS X, and you replied that you would make an
>>attempt as soon as the next version was finished.
>>
>>Version 0.17a has the perfect strength as a sparring partner for my own
>>engine.  The last time I tried, my engine lost a 100-game match by 47.5-52.5.
>>An engine significantly stronger than Ruffian would just be too depressing
>>to play against, I'm afraid.  :-)
>
>Wow, either Gothmog was very lucky or something was broken in ST's setup or
>Gothmog is a lot stronger than I think.
>:)

There was certainly some luck involved.  My other results with Gothmog 0.4.7
have not been quite as good.  Here are the results of the matches I have
played so far:

Gothmog 0.4.7-Yace Paderborn:  40-60
Gothmog 0.4.7-Pepito 1.59: 42.5-47.5
Gothmog 0.4.7-SmarThink 0.17a: 47.5-52.5
Gothmog 0.4.7-AnMon 5.21: 56-44
Gothmog 0.4.7-Phalanx XXII: 71.5-28.5

Tord



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.