Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Control extensions

Author: Bas Hamstra

Date: 13:23:03 09/25/99

Go up one level in this thread


Hello Bob,

I just wanted you to know it works ok. Very flexible.

I am still having a problem with the mate (nullmove returns mate) extension,
though. That thing never has worked very good for me. Do you use it?

Currently I am trying this settings, after some fiddling:

(1 ply = 100)

- OneReplyMate = 120
- Check = 60
- Recapture = 30
- NullReturnsMate = 80

I don't extend when Depth > 2*RootDepth, except for OneReplyMate, which doesn't
seem to cause problems.

I somehow feel more could be done with the NullReturnsMate. Now it solves a few
wac's extra for me, with sometimes quite some overhead. It seems to be able to
spot forced lines, but I don't think my program profits enough from them. Are
there any tricks?

Regards,
Bas Hamstra.







On September 24, 1999 at 10:15:27, Robert Hyatt wrote:

>On September 24, 1999 at 08:14:05, Bas Hamstra wrote:
>
>>What are good ways to control extensions? I like the idea of extending part of a
>>ply, but am not sure how to implement it.
>>
>>Suppose you have a depth 5 to search. Then if you wanted to extend 1/2 ply then
>>there are now 2 available extensions. You extend here. Next ply: depth=4. How
>>many available extensions? At first I thought 4/2 = 2, but already used 1, so 1
>>left. But I think this is not quite correct to subtract "history" extensions
>>from the available extensions this ply. But how else???
>>
>>
>>
>>Regards,
>>
>>Bas Hamstra.
>
>
>First, for simplicity, lets count plies by 100.  So a five ply search starts
>with a depth of 500.  For every ply you advance into the tree, you decrement
>this by 100.  So it works as always and the *100 doesn't do a thing, yet.  But
>now we extend by 3/4 ply, say (for some reason at the root of the tree) so our
>depth is now 575.  This does nothing (yet) as when depth drops to < 100, we
>treat it as zero.  So we are still doing a 5 ply search.  But _if_ we extend
>by a fraction of a ply at another point in the tree, 575 + 75 = 650 and _now_ we
>are really doing a 6 ply search, not 5.  But notice how it took two extensions
>at two different plies, to make this happen?  Which lets you sort of say "if
>something interesting happens at two different nodes in the same path, then I
>will extend that deeper than if something interesting only happens one time in
>a path..."



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.