recapture extension question?

[Daniel]

Geschrieben von: / Posted by: Daniel at 23 January 2004 07:09:37:

Hi
Do I have to extend losing captures in search?
Extending only winning captures seems to make my program
a little bit faster.I have read in Ed Shroeder's paper Rebel
uses recapture ranges and extends all captures(winning and losing).
My question is if I reach the same _material as the root_material by
losing,say QXN,do I have to extend this ply.
regards
Daniel
P.S winning also includes equal captures

[Uri Blass]

Geschrieben von: / Posted by: Uri Blass at 23 January 2004 10:00:09:
Als Antwort auf: / In reply to: recapture extension question? geschrieben von: / posted by: Daniel at 23 January 2004 07:09:37:

Hi
Do I have to extend losing captures in search?
Extending only winning captures seems to make my program
a little bit faster.I have read in Ed Shroeder's paper Rebel
uses recapture ranges and extends all captures(winning and losing).
My question is if I reach the same _material as the root_material by
losing,say QXN,do I have to extend this ply.
regards
Daniel
P.S winning also includes equal captures

You do not have to extend if it does not help you.
The decision what to extend is your choice.
It is impossible to give a general reply because it is dependent in the engine but I guess that basically extending bad moves is a bad idea and it is better to prune them.
I am sure that conditions when not to extend checks can also help and the problem is to define the right conditions when never extending checks that sacrifice material does not seem to me a good idea.
Uri

[Tord Romstad]

Geschrieben von: / Posted by: Tord Romstad at 23 January 2004 12:03:13:
Als Antwort auf: / In reply to: Re: recapture extension question? geschrieben von: / posted by: Uri Blass at 23 January 2004 10:00:09:

Hi
Do I have to extend losing captures in search?
Extending only winning captures seems to make my program
a little bit faster.I have read in Ed Shroeder's paper Rebel
uses recapture ranges and extends all captures(winning and losing).
My question is if I reach the same _material as the root_material by
losing,say QXN,do I have to extend this ply.
regards
Daniel
P.S winning also includes equal captures

You do not have to extend if it does not help you.
The decision what to extend is your choice.
It is impossible to give a general reply because it is dependent in the engine but I guess that basically extending bad moves is a bad idea and it is better to prune them.
I am sure that conditions when not to extend checks can also help and the problem is to define the right conditions when never extending checks that sacrifice material does not seem to me a good idea.

Yes. What works well in one program often does not work at all in another
program. The only way to find out is to try it out and play an enormous
number of games. It's boring, frustrating and time-consuming, but unfortunately
necessary.
For me, recapture extensions in all forms are bad. I've tried them several
times, but my test results have always been better without them.
A reasonable compromise is to always extend checks, but not always equally much.
My check extensions vary between half a ply and a full ply.
Tord

[Daniel]

Geschrieben von: / Posted by: Daniel at 23 January 2004 16:24:22:
Als Antwort auf: / In reply to: Re: recapture extension question? geschrieben von: / posted by: Uri Blass at 23 January 2004 10:00:09:

Hi
Do I have to extend losing captures in search?
Extending only winning captures seems to make my program
a little bit faster.I have read in Ed Shroeder's paper Rebel
uses recapture ranges and extends all captures(winning and losing).
My question is if I reach the same _material as the root_material by
losing,say QXN,do I have to extend this ply.
regards
Daniel
P.S winning also includes equal captures

You do not have to extend if it does not help you.
The decision what to extend is your choice.
It is impossible to give a general reply because it is dependent in the engine but I guess that basically extending bad moves is a bad idea and it is better to prune them.
I am sure that conditions when not to extend checks can also help and the problem is to define the right conditions when never extending checks that sacrifice material does not seem to me a good idea.
Uri

I have just quoted the code where my extensions are located as a "VERY Danagerous" area,especially after reading what was posted in CCC.Using the CCC search engine(very useful tool for programmers IMO) I have discovered that
1)Check extensions are really dangerous to mess up with
suggested methods to limit
.not extending near the horizon
.not extending the first capture -> I just tried this one.
Search is fast and can solve some test positions quick but
not prominsing in actual games.

2)recaptures are not as important as they seem ,i have decided to extend only
winning captures which restore material.Some people even said it isuseless.
And a lot more interesting topics on other recaptures.
I am going to experiment with them very carefully( i mean VERY).
Thanks (both Uri and Tord) for your suggetions.
regards
Daniel

[Bas Hamstra]

Geschrieben von: / Posted by: Bas Hamstra at 23 January 2004 21:18:30:
Als Antwort auf: / In reply to: Re: recapture extension question? geschrieben von: / posted by: Tord Romstad at 23 January 2004 12:03:13:

For me, recapture extensions in all forms are bad. I've tried them several
times, but my test results have always been better without them.

I also have problems with recap extensions. I have very mixed results. They do not improve my program in testsuites and selfplay. However when I play Crafty in blitz it suddenly makes a huge difference. So I retried them for a while for example in the Leiden version of Tao. I am not quite satisfied with it in longer timecontrols. I am now experimenting with alternative forms, some of which seem to improve tactics somewhat. My impression is that it might be a win in blitz and a loss in long time controls.
Bas.