Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: For Ed: thanks for a quick answer - explanation

Author: Ed Schröder

Date: 03:14:59 09/26/03

Go up one level in this thread


On September 26, 2003 at 05:01:05, Werner Schuele wrote:

>Hi Ed, here in other words:
>when you would try to make a rebel.eng similar to those of Thorstens, but
>without these singular extensions, which parameters do you use instead?
>
>Or is it ok, if we use these parameters together?

In general: one must be very careful not adding too many extensions, too many
extensions have the nasty habit to perform very well at the early plies (blitz)
but may become counter productive at deeper depths (tournament time control).

Therefore there must be some harmony in its use. The right use of extensions is
already a major pain for the engine programmer, finding the right harmony seems
almost mission impossible for the interested user. Of course Thorsten will
disagree with me :-) but he has some right to speak as I consider his knowledge
about Rebel quite amazing.

However here are some general guide lines for using the extension parameters in
the upcoming Rebel 12:

When you are using Singular Extensions (SE) it is wise not to add more extension
parameters because SE is a costly business, Rebel typically will calculate one
ply less deep.

As a general rule: do not slow down the depth of Rebel more than a factor of 2
adding extension parameters, average spoken of course.

Let me describe my findings using Rebel12's extension parameters, experiences
are based on using test positions and playing Rebel-Rebel engine matches, thus
no hard guarantees.

[Pruning = MISC_28]             * More check extensions

Gave only a minor improvement. Typical slow done in depth about 20%.


[Pruning = MISC_30]             * Only one escape from check extension

Equal in strength, typical slow down about 20-25%.


[Pruning = MISC_37]             * New extension

Gave a huge elo jump playing blitz games, it has to be seen if this extension
also works at longer time controls and against other programs, maybe it just
kills the old Rebel only. But this extension is something you could focus on.
Typical slow down about 10-15%.


[Pruning = MISC_38]             * New capture extension

Gave a very small improvement, slow down 20-25%.


[Pruning = MISC_44]             * New endgame extension

Gave also a very small improvement, slow down in the end-game 5-10%.


[Pruning = MISC_46]             * 6th rank extension

Gave good results at blitz, not so good at longer time controls. Perhaps there
is an improvement using the right sub-parameters. Estimated slow down 10-15%.


[Pruning = MISC_47]             * King Safety extension

Equal in strength, slow down 5%, nice to have.


[Pruning = MISC_56]             * Nullmove extension

Gave a +8 elo improvement at blitz, haven't test it yet at longer time controls.
Slow down is not much, 3-5%


[Pruning = MISC_57]             * extension if dangerous pin

It seems I have forgotten to document its results. By head: it gave no
improvement but wasn't that bad also.

Happy tuning...

Ed




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.