Author: Enrique Irazoqui
Date: 02:37:10 11/02/99
Go up one level in this thread
On November 01, 1999 at 18:39:40, Heiko Mikala wrote:
>Hi Enrique!
Hi Heiko :)
>On November 01, 1999 at 12:32:35, Enrique Irazoqui wrote:
>
>
>
>First of all, since this is a heated discussion and I burnt my mouth once
>already, I'd like to point out that this post is not meant to attack anybody,
>just to comment on an interesting part of the discussion.
>;-)
>
>
>>I checked for differences in Crafty as a winboard engine and as a native engine
>>for Fritz, and I could not detect any, except for winboard running about 2%
>>faster in NPS than Fritz. I checked this with games and test positions: same
>>moves, same PVs. Below I copy some.
>
>I think the much more important question is, how does Fritz influence Crafties
>behaviour in between the moves? I'm sure, that there is no difference in
>analysing fixed positions, but what happens during a game?
As I said a couple of days ago, only Bob and Mathias Feist can answer to this. I
have no idea about the inner guts of Crafty and Crafty for CB, so the only thing
I could do is to compare positions and games with Wcrafty and Crafty for Fritz.
I had never seen Winboard until a couple of days ago, when Djordje sent it to me
because I wanted to check for these possible differences. By the way, Winboard
may be useful, but what a relief to go back to Fritz!
So I checked the best I could thnk of and I saw no differences "as far as I can
tell", which I don't know how far it is. For instance, Crafty for CB does not
clear hashtables as it did with 16.1 a year ago. But what happens for example
with the hashtables? I see that in Winboard the Crafty.rc file has a separate
entries for "hash" and "hashp", with a ratio of 3:1 in the files I saw. In
Crafty for Fritz there is only one generic "hashtable size" and I don't know if
Mathias made it work with this 3:1 ratio. I don't know either to what extent it
matters. So the best I could do, I think, was replaying a game with Wcrafty
16.18, trying to follow the times used by Tiger. Not easy at all at blitz, but I
think I managed and Wcrafty played the same moves. Also the positions I posted
yesterday were solved the same way. So "it seems" that there is no difference.
Since Bob escapes from Windows and Fritz as fast as he can manage, only Mathias
Feist could tell us more about all this.
To me, it was interesting to see the effect of PB off, but I am the only one
that seems to care. :(
>My own observation is, that Crafty as a winboard engine, run through the
>winboard adapter in Fritz, plays as much as 100 elo points weaker than the same
>Crafty version as a "native" Fritz engine. I know that you used the native
>version, but do we know, if there is a difference between the native Fritz
>engine and the real Crafty, running as Winboard engine in Winboard? What, if
>there is again another 100 or 50 points difference?
See above.
>I don't know, if there is a difference, but there happened one thing today that
>made me wonder again: during the last months I have run a lot of blitz
>tournaments on my computer, all g/5, all using Fritz as the interface. In my
>rating list, generated by Fritz, which contains more than 2200 games by now,
>Crafty (Fritz native engine) and Little Goliath 2000a are roughly equal in
>strength. Today I had a look Frank Quisinkys homepage again, who, as you surely
>know, runs large tournaments using Winboard. But in his latest rating list (on
>the news ticker page), which also contains about 2000 games IIRC, Crafty
>(Winboard engine) is about 100 points ahead of Little Goliath...
It seems ("on my computer", singular) that you played in eng-eng form within
Fritz. I don't trust this way of playing. If Frank used 2 computers, his
results are more reliable.
>I don't know, how to answer this question without investing a lot of time, but
>at the moment I'm running a g/60 tournament on two computers and maybe I'll
>enter Crafty as a real Winboard version after I have finished the current
>tournament. This would of course not answer the question, if there is a
>difference between the real Crafty and the Fritz engine, but it could tell us a
>little bit about how strong Crafty is compared to the commercials. It would have
>to play against F6, H7.32, CM6000, Rebel Century, MChess Pro and Genius 5.
>Didn't you do something like this last year, including Crafty?
I entered a crippled Crafty on my 550 games tournament, but it ended up in a
royal mess. It was Crafty 16.1 for Fritz and this particular version had
problems, like clearing hashtables with new move, so its performance was
probably not realistic.
I think that a necessary condition to do well in comp-comp is to be a very fast
finder, not sufficient, but necessary. So let me make a prediction for your
tournament, assuming that you will play a large number of games on 2 computers:
CM6K, F6 and H732 will be on top; Rebel and Crafty will end up half way, and
Mchess and Genius will be far behind.
I am not sure at all that being a fast finder is as important in human-computer
games and that comp-comp results can be extrapolated to performance against
humans. So about trying to find out through comp-comp "how strong Crafty (or any
program) is compared to the commercials" in human-comp, I am a bit skeptical. I
think that comparing comp-comp with human-comp is entering unknown territory.
Enrique
>Greetings,
>
>Heiko.
>
>
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.