When engine-engine games are played on the Fritz GUI and a middlegame draw results, is it the engines that 'agree' to the draw, or is it Fritz that imposes one?
It's the GUI, and sometimes it does it wrongly, that's a reason I don't use it for engine matches at all.
You can select Draw:Never in the game options (under Tools/Options).
I like any option which helps to avoid the nightmares of unneccessary long games. Almost all extremely long games are draws the engines don't understand. I give priority to get rid of that as good as possible, even if one or two half points in a long(!) match or tournament are wrong. The statistical uncertainty is bigger anyway.
Another option is the move limit for games. If I use that, I review each game which has reached that limit. Almost all of the extreme long games are clear draws (typical example: pawns and bishops on different square color). Fritz does not assign a result to such games, so I need to adjudicate them anyway. Arena assings draw but I always reviewed these long draws, replacing them with wins in very seldom cases. Also, there may be cases where the positions seems in fact winable, but the last 100 moves clearly show that the engine on the better side wasn't capable, under the conditions each. IOW in some cases I do not only look at the end position, but also what happened before, to find a fair adjudication.
That takes only a couple of minutes, but can save up to hours in a long test.
I like any option which helps to avoid the nightmares of unneccessary long games. Almost all extremely long games are draws the engines don't understand. I give priority to get rid of that as good as possible, even if one or two half points in a long(!) match or tournament are wrong. The statistical uncertainty is bigger anyway.
Another option is the move limit for games. If I use that, I review each game which has reached that limit. Almost all of the extreme long games are clear draws (typical example: pawns and bishops on different square color). Fritz does not assign a result to such games, so I need to adjudicate them anyway. Arena assings draw but I always reviewed these long draws, replacing them with wins in very seldom cases. Also, there may be cases where the positions seems in fact winable, but the last 100 moves clearly show that the engine on the better side wasn't capable, under the conditions each. IOW in some cases I do not only look at the end position, but also what happened before, to find a fair adjudication.
That takes only a couple of minutes, but can save up to hours in a long test.
If it's a repetition/50-move/insufficient material draw, the engines agree. Otherwise it's always the GUI making the decision (with the exception of Winboard-native GUIs, but Fritz isn't one of those).
> Otherwise it's always the GUI making the decision
It's specially annoying when the GUI adjudges games as draws when one side is clearly winning just because one engine isn't making progress score-wise and the other always shows 0.00 (9.00 0.00 - 9.00 0.00 - 9.00 0.00, draw agreed??)
It is not hard to fix.
programmers can simply not use their normal evaluation in the pv that they give when they play under Fritz gui.
They can simply calculate evaluation=0.01*number of ply to play when the engine always has the advantage.
possible games without book may start in the following way:
1.e4 0.01/14 1...e5 -0.02/18 2.Nf3 0.03/15 2...Nf6 -0.04/19
hopefully stupid Fritz is going to adjudicate nothing if the engines do something like that.
Uri
programmers can simply not use their normal evaluation in the pv that they give when they play under Fritz gui.
They can simply calculate evaluation=0.01*number of ply to play when the engine always has the advantage.
possible games without book may start in the following way:
1.e4 0.01/14 1...e5 -0.02/18 2.Nf3 0.03/15 2...Nf6 -0.04/19
hopefully stupid Fritz is going to adjudicate nothing if the engines do something like that.
Uri
Just set resign never and draw never in the gui.
I guess that you are right.
I have only Fritz8 and I remember that this strategy worked for me when I ran engine-engine match under that gui but I assumed that if people complain about the problem it may be possible that there is no way to fix draw adjudications because of a bug in later versions of Fritz.
Uri
I have only Fritz8 and I remember that this strategy worked for me when I ran engine-engine match under that gui but I assumed that if people complain about the problem it may be possible that there is no way to fix draw adjudications because of a bug in later versions of Fritz.
Uri
Are you claiming that if the gui is set to draw never, the server will not draw games that have a sequence of 0.00 moves? I don't think that's right...
Alan
Alan
> Just set resign never and draw never in the gui.
I had draw set to never when this happened :(
I've never seen that, though. The draws of the Fritz GUI which I have seen are either because both engines show 0.00 for some moves or because other things like repetition or 50 moves rule ascend on the horizon.
In the contrary I find that the Fritz GUI tends to prolong some games nearly unendingly, just because one of the two engines doesn't want to accept that the position is dead draw and stubbornly decides to believe it has an 4 centipawns advantage...
In the contrary I find that the Fritz GUI tends to prolong some games nearly unendingly, just because one of the two engines doesn't want to accept that the position is dead draw and stubbornly decides to believe it has an 4 centipawns advantage...
Well, the situation that prompted the question was this: I set Deep Shredder 11's contempt to 1.25 while playing blitz 5 + 1 with Rybka 2.2.3a and I started to get quite a few draws (relatively speaking). I presume that Shredder was 'offering' the draws, but what surprised me was that Rybka 'accepted' these presumed draw offers when there was still plenty of play in the position. I didn't record the engine's evaluations, but I'm pretty sure Rybka's self evaluation was slightly positive at least for some of the draws. I'm going to try this again and record the evaluations at the time of the draws.
<---- interested in the result.
there is also the "290 moves rule" in Fritz that stops games after a lot of moves... quite strange :)
Where do you guys get this info? I don't play engine/engine matches now but when I played auto232 matches the GUI did not control the draw! It had to first be offered by one engine and then accepted by the other. If the second engine rejected the draw the game continued. The "engine offers draw" message showed up on the screen and was often refected by the second engine. Was this all a farce put on by the GUI ?
I think that the posters do not talk about auto232 matches but about games under the fritz gui in a single computer with ponder off.
I believe that most people test in this way(CEGT and CCRL are with ponder off)
Uri
I believe that most people test in this way(CEGT and CCRL are with ponder off)
Uri
Hello Uri,
I understand that but why would they make the GUI in charge of draws in engine/engine matches and not in auto232 matches? If the engines are capable of this decision in auto232 then they should also be capable in engine/engine games.
Regards,
Jim Walker
I understand that but why would they make the GUI in charge of draws in engine/engine matches and not in auto232 matches? If the engines are capable of this decision in auto232 then they should also be capable in engine/engine games.
Regards,
Jim Walker
I think the GUI makes this decision based on the evaluations. That is why clear draws can end in games with 200+ moves, because one, or both engines do not understand it and never provide a 0.00 eval. The GUI will not end a game as a draw if one engine disagrees (except draw by the rules, like repetition etc.).
Everyone should take a look at: Tools/Options, Game. There, draw agreements and resignements can be adjusted.
Everyone should take a look at: Tools/Options, Game. There, draw agreements and resignements can be adjusted.
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill
