Laskos wrote:
At 5x time control the White performance increased from 62% to 71%, and this, 5x longer time control, is still short. My hardware is weak compared to Pal's. It is probable that at LTC & strong hardware the White performance is >80% in this position.
Pal, my opinion is that it would be better to remove the games 85 & 86 (and the position).
Kai
Even worse, here is the summary for different TC, engines on 1 core:
2s + 0.2s ____62% White perf.
10s + 1s ____71% White perf.
30s + 3s ____76% White perf.
Your TC is 50 times longer than 30s + 3s, your hardware 10 faster, totally a factor of 500. With that steady pace of increasing White performance with time, I bet the games 85 & 86 were >95% won for White from the opening position, for both Houdini and Rybka. These games should be removed (the position too).
Laskos wrote:Even worse, here is the summary for different TC, engines on 1 core:
2s + 0.2s ____62% White perf.
10s + 1s ____71% White perf.
30s + 3s ____76% White perf.
Your TC is 50 times longer than 30s + 3s, your hardware 10 faster, totally a factor of 500. With that steady pace of increasing White performance with time, I bet the games 85 & 86 were >95% won for White from the opening position, for both Houdini and Rybka. These games should be removed (the position too).
Kai
Thorough analysis, thanks!
All the more surprising that Rybka nearly managed to draw with black.
Unfortunately, with 4,000 positions in the file, there is no way for me to assess them before they are used. I had thought that they were all tested since Bob Hyatt put them on his site, and apparently used them for testing himself (I think).
I do not think that I should start to "remove results" based on the opinions of whether a position is "balanced enough". It is too subjective (even though you did test) and would only cause problems. If a few positions are truly favorable to one side or the other then both getting a win\loss seems ok.
The alternative would be difficult to manage since the GUI does not take these "removals" into account, so a 100 game match would end and you may have only 85 or 90 games played. Then if you restart a match to get the extra games, some "random" position that was used could be used again. I would also have to remember to change the W-L numbers by hand as I report them to remove the "adjustments", an error-prone process.
That would be interesting. I do have a few "older" ones laying around but I do not have another pc equal to the one that I am using for the match. Maybe in the future if I have 2 that are "close enough" I will give it a try.
PawnStormZ wrote:
The alternative would be difficult to manage since the GUI does not take these "removals" into account, so a 100 game match would end and you may have only 85 or 90 games played.
Hi Pal,
current ChessGUI does not pick duplicate openings.
You can interrupt the match, then send me pgn and epd to weed out played positions, then upgrade to current ChessGUI before continuing the match.
I did not realize that Pal used ChessGUI...
Then why are his PGNs legible by ChessBase - including the Evaluation Profile, while Martin's (Thoresen) are not?
