Author: Christophe Theron
Date: 20:15:01 11/09/01
Go up one level in this thread
On November 09, 2001 at 22:20:10, Thomas Mayer wrote:
>Hi Christophe,
>
>>> In most rating lists, Grizzly is around 1800-1900 - this new version is a
>>> little bit stronger, but I doubt that it is over 2000. Does this mean that
>>> Chess Tiger on Palm is also clearly under 2000 ?
>
>
>First of all: Sorry, Christophe, I have read my statement again and it sounds
>more offensive then it was ment. (Talking about feeling how not native langueges
>function - there is always a problem...)
>I am a little bit a cheeky fellow and that was the idea behind it - somehow it
>is unfair to compare an Athlon 1333 with those Palms, especially in Blitz
>games... In those fast games it is clear that the depth is the main issue and I
>believe that Chess Tiger on Palm gets clearly outsearched by Grizzly. This might
>change with longer time controls, but the speed difference between the
>processors is still extrem.
>But on the other hand I was amazed - I did not thought that one of those engines
>would even had a chance against Chess Tiger, in fact I was quite sure that not
>my own engine would have any chance against it. But it's definitely different in
>computerchess then in human chess, that's something I am totally sure about it.
>I think when we take a player rated around 2000 and let him play against Grizzly
>and Chess Tiger he would get the better result against Grizzly and ChessTiger
>might give it a much harder task. But that's just a guess...
>
>>1) How can you base any strength evaluation on 3 games only?
>
>Well, I can't - but I am quite sure that it may go on that way... maybe not as
>clear, ChessTiger might get some wins either, but I believe that Grizzly on that
>fast Athlon will have his head above. (I can not test myself, I don't have a
>Palm...)
>
>>2) You say that Grizzly is around 1800-1900 "in most rating lists". But Grizzly
>>is not on the SSDF list. Your post suggests that the SSDF ratings cannot be
>>compared to FIDE ratings. Why don't you start by assuming that the rating lists
>>in which Grizzly appears cannot be compared to the SSDF list, and that the
>>ratings cannot be compared? So basically we have no valid rating for Grizzly.
>
>We have, because included in those lists are some programs that also have a
>rating at the SSDF (e.g. Little Goliath, SOS, Crafty) - so at least the lists
>are compareable.
No they are not unless in the rating lists you are talking about Crafty and the
other known programs have exactly the same rating as in the SSDF list.
I do not know what lists you are talking about, but the rating of the chess
servers for example are totally useless here.
> So we HAVE a rating for Grizzly. If that is not enough for you,
>I will let it play against some more engines mentioned on the SSDF list. Against
>the strong engines it will not get much on compareable hardware - the given
>rating might be even lower then what I have said above.
>That's the point I want to say, from human sight I believe that the difference
>between the nowadays tops and the older days dedicated computers is MUCH to high
>on the SSDF-list, because those old computers suffer to much from the missing
>depth which does not help as much in human games.
>And when we start now to let the old irons play against the top engines, the
>difference might even grow (and I am absolutely sure that it will grow). So the
>ratings on the SSDF are somewhat unclear and the induction to compare them with
>FIDE-ELO is problematic, so my proposal is to give the SSDF-rating a different
>start value that this comparisson stops.
>
>> The valid way to evaluate the strength of Chess Tiger for Palm is to let it
>> play against computers that have a SSDF rating.
>
>if that is wanted I will try to get one and let it play against Fritz 7, Junior
>7, ChessTiger 14, Rebel Century 3, Shredder 5, Gandalf 4.32 etc... Do you think
>it will earn 2100 ??? In fact I am quite sure that it HAS around 2100 ELOs, but
>in comp-comp match depth is more important then against humans and those
>differences get to big on the list.
If you match it against too strong opponents, you'll get a mathematically high
margin of error. So that will not work.
You need to let it play against computers/programs listed in the SSDF which have
a rating close to the estimated rating of Chess Tiger for Palm.
>> Other methods are prone to a huge margin of error.
>
>The margin of error will be even bigger when I test it like it is described
>above.
Right.
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.