Author: Christophe Theron
Date: 14:34:29 02/27/01
Go up one level in this thread
On February 27, 2001 at 16:40:09, Uri Blass wrote:
>On February 27, 2001 at 12:47:23, Christophe Theron wrote:
>
>>On February 27, 2001 at 01:06:19, Uri Blass wrote:
>>
>>>On February 26, 2001 at 23:13:26, Christophe Theron wrote:
>>><snipped>
>>>>The question of the branching factor is a crucial one indeed, and as far as I
>>>>know it is the only reason that could explain that a program needs faster
>>>>hardware or longer time controls.
>>>>
>>>>For example, Genius has an horrible branching factor, which is the reason why it
>>>>is still good at blitz (well at least on slower computers) and so bad at long
>>>>time controls on current hardware.
>>>>
>>>>I think that it should be possible to back up claims that a given program needs
>>>>faster hardware or longer time controls by measuring its branching factor. If a
>>>>program has a much better branching factor that the other ones, then it is a
>>>>strong indication that it will be superior as time controls and/or processor
>>>>power increases.
>>>>
>>>>Any volunteers?
>>>
>>>If you get a smaller branching factor from a right pruning idea then you made an
>>>improvement for long time control but it is also possible that you got a smaller
>>>branching factor by pruning some logical moves and in this case the smaller
>>>branching factor is counter productive.
>>
>>
>>Of course. You know, it's the basic fight of the chess programmer!
>>
>>
>>
>>>It is also possible that you get a bigger branching factor from adding some
>>>extensions and the extensions may be productive at long time control.
>>
>>
>>Why should extensions be productive more at longer time controls?
>>
>>It could very well be the opposite.
>
>I agree that it can be also the opposite but I can imzgine extensions that are
>productive only at long time control and my example of playing against yourself
>and learning is an example.
You first have to prove that your extension is productive.
>>>I think that the main problem of Ganius(at least Genius3) is not enough
>>>extensions at long time control and not an horrible branching factor.
>>>
>>>I know that Genius3 never extends more than 12 plies.
>>
>>
>>This limit is absolutely no problem in practice.
>
>It is a problem at long time control because there are cases when you need to
>see a lot of plies in some selective search and brute force depth+12 that is
>often 25 or 26 is not enough.
But Genius uses a SEE and mate detector, which means that it evaluates the
remaining captures even in the deepest ply of its QSearch. In pratice it can add
several more plies of lookahead.
You think it stops at depth+12, but it is actually able to see the captures
sequences happening beyond that point.
I really don't see it as a problem.
>>>Genius3 from my experience has typical branching factor of 3-4 even at long time
>>>control and my experience with hiarcs says that the branching factor of hiarcs
>>>at long time control is clearly bigger.
>>
>>
>>Genius takes more and more time to reach successive ply depth. At the start ot
>>the search it is as fast as other programs (and even faster), but from ply 5-6
>>(what other programs typically call 8-9 plies), it begins to be very slow.
>
>This is not my experience with Genius3.
>
>I remember that the typical branching factor at the middle game was between 3
>and 4(usually closer to 4 at long time control but not 6 or 7 that is a typical
>branching factor of hiarcs at long time control).
>
>It is the case in the middle game and in the endgame it could be even close to
>2.
A branching factor between 3 and 4 is a bad one by today's standards.
Christophe
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.