Author: Christophe Theron
Date: 12:24:00 04/22/01
Go up one level in this thread
On April 22, 2001 at 14:48:56, Robert Raese wrote:
>On April 22, 2001 at 13:51:44, Christophe Theron wrote:
>
>>On April 22, 2001 at 13:24:19, eric guttenberg wrote:
>>
>>>Mike,
>>>
>>>I don't know what selsafety is, though I suspect it may have something to
>>>do with king safety. It is found in the Chess Tiger notepad which is in
>>>your Rebel file. The default value is 6; several people have changed the
>>>value to 12 and have reported improved winning percentages. Christophe
>>>Theron responded to one such report by saying he would take a look at
>>>it in his work on the Tigers. I don't know if there is any consensus
>>>from Tiger users as to whether using the Selsafety value of 12 is
>>>better or not, but it does appear possible.
>>>
>>>
>>>eric
>>
>>
>>
>>You will find the "SelSafety" line in a file named CT.INI, which is in the
>>program's installation directory (typically C:\Program
>>Files\Rebel\Rebel-TigerII).
>>
>>It has nothing to do with king safety.
>>
>>It is about Selectivity Safety. Higher values give a better safety to the
>>selective algorithm (at the cost of slowing the program down). Lower values
>>increase the effect of the selective system, speed the program up, but also
>>increase the risk of selectivity errors (missing an obvious tactical shot for
>>example).
>>
>>The default value of 6 was optimal at the time we released Chess Tiger 13. I
>>have not tried to change this setting in Chess Tiger 14, so it is not impossible
>>that given the other changes in the engine another default value would perform
>>better.
>>
>>I hope I will get enough feedback on very long matches (400 games or more) to
>>figure this out.
>
>how does the time control for these test games influence the result?
>
>usually i play tiger at ICC with time control of 5 minute blitz. if i want to
>know if the +12 setting is stronger at game in 5 + 3, is it necessary to test at
>that time control, or can information about that be extrapolated from the tests
>at longer time controls?
I think you can safely extrapolate in this particular case, due to the nature of
the parameter in question.
But we are not absolutely sure about the efficiency of the new parameter value
at this time, so don't jump on it blindly.
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.