Page 4 of 5
Re: Hannibal 1.7
Posted: Wed Aug 10, 2016 10:01 am
by Edsel Apostol
IWB wrote:Edsel Apostol wrote:IWB wrote:Hello Edsel,
I started a test already. If it gets into the TOP 16 it will throw out the old horse Naum
I remember Hanibal 1.5 having some ponder issues which causes a lot of time losses. The first 20 games of 1.7 are fine ... I am crossing fingers!
Thanks for the version
Ingo
Hi Ingo, we have fixed the issue with ponder games on the Shredder UI. We'll be looking forward to see it in your list.
Hello Edsel,
It is not completly gone. In 2014 games I have 7 losses on time. That is nothing to worry about but it is still there.
Do you allways keep some time on the clock left or do you play in the last fraction of the last second?
Besides that, the performance until now is briliant!
Ingo
We subtract a default 1 second ("Time Buffer" UCI option) to the available time before we calculate the maximum time when the engine should return a move, so it shouldn't lose on time. My testing with 10s + 100ms TC doesn't have any time losses but it is not ponder on games and it uses cutechess, so it's probably a UCI implementation issue when using the Shredder interface, like not replying to isready or returning an empty bestmove. I'll investigate further.
In the meantime, hopefully it retains the good performance until the end of the matches!
Re: Hannibal 1.7
Posted: Wed Aug 10, 2016 10:51 am
by IWB
Edsel Apostol wrote:
We subtract a default 1 second ("Time Buffer" UCI option) to the available time before we calculate the maximum time when the engine should return a move, so it shouldn't lose on time. My testing with 10s + 100ms TC doesn't have any time losses but it is not ponder on games and it uses cutechess, so it's probably a UCI implementation issue when using the Shredder interface, like not replying to isready or returning an empty bestmove. I'll investigate further.
In the meantime, hopefully it retains the good performance until the end of the matches!
After 2230 games it is on 8 time losses now. With your current average performance of 45% Hannibal might have lost 4 point of 998 ... that is a fraction of ONE Elo point. Again, nothing to worry about regarding the rating (would be different in a torney
). The previous version lost massive Elo points with a similar bug.
All I can see is that Hannibal never gets out of the ponder. The opposite engine moves after 0s - 22s (min and max of the 8 games) and Hannibal simply continues to ponder. I would assume is somehow misses the opponents move. It happenes with white and black. Btw, it is not always the same opponent, it is actualy 8 different ones now.
Ingo
Re: Hannibal 1.7
Posted: Wed Aug 10, 2016 10:55 am
by IWB
Guenther wrote:IWB wrote:
It is not completly gone. In 2014 games I have 7 losses on time. That is nothing to worry about but it is still there.
Do you allways keep some time on the clock left or do you play in the last fraction of the last second?
Besides that, the performance until now is briliant!
Ingo
According to your site you play with +3s increment, is that right?
Then it shouldn't really lose on time.
Is there any pattern to those 7 time losses, e.g. resolving fail lows
in already bad positions, or very drawish endgames near the 50 moves draw?
Guenther
Yes with 3s increment.
It is happening in all phases of the game. 5 moves in the game (with lot of time on the clock) and towards the end. Most of the time in equal positions but I have 2 lost games and one theoretical won game (which then was lost) ...
It is just not getting out of ponder after the opponents move.
Ingo
PS: I only noted this because of the massive problems with the previous version, without that I would not have even looked, maybe at the end of the tourney... . I know some other engines with similar problems when playing high numbers of ponder ON games. Ponder isn't trivial but part of the game - unfortunately not tested very often ...
Re: Hannibal 1.7
Posted: Wed Aug 10, 2016 11:27 am
by Frank Quisinsky
Hi Edsel,
wow, very friendly!
Thanks!!
For the Moment I am looking in test-results by others.
I think I can start with Hannibal in around 4 weeks.
Really a nice time in computer chess.
Elo average from TOP-50 today is ... 2 8 4 3 !!
Speaking from average!
If you say it to me for 15 years ... 2 8 4 3 ... never I believe that.
Good luck and again ... have many thanks.
Long time not updated engines are an event for me because we lost enough strong engines in the last years.
Best
Frank
Re: Hannibal 1.7
Posted: Wed Aug 10, 2016 11:28 am
by IWB
Just as a sidenote:
In the current 26400 games of the top 16 (without Hannibal) I have 38 time losses. Even the top 3 engines like SF, Komodo and Houdini had a few ...
It happens - but Hannibal is slightly elevated!
Ingo
Re: Hannibal 1.7
Posted: Wed Aug 10, 2016 1:39 pm
by cdani
Just to comment it, I leave unused upto 3 x increment time, so should be impossible that Andscacs loses on time. If it happens is something of the operating system.
Re: Hannibal 1.7
Posted: Wed Aug 10, 2016 1:58 pm
by IWB
cdani wrote:Just to comment it, I leave unused upto 3 x increment time, so should be impossible that Andscacs loses on time. If it happens is something of the operating system.
I dont have a single time loss for Andscacs 0.871!
This is the chart out of 26400 games, 3300 games per engine:
Komodo 10.1, Texel 1.06, Critter 1.6a = 1 loss on time
Gull 3 = 2 losses on time
Fritz 15 = 3 losses on time
Nirvanachess 2.3, Equinox 3.30, Stockfish 7 = 4 losses on time
Chiron 3 = 18 losses on time
But even with 18 losses Chirons average performance is ~40% that would be 7 more points to the 1307.5 ... that might be 3 Elo but my error bar is 10 Elo wide ... for all the others it doesn't matter at all.
And of course, loosing on time belongs to the game!
Ingo
Re: Hannibal 1.7
Posted: Wed Aug 10, 2016 6:05 pm
by MikeGL
supersharp77 wrote:Edsel Apostol wrote:MikeB wrote:
Engine error Hannibal 1.7w32.....(not valid w32 engine)
Thx for the attempt...AR
Same here. I am running it on a native Win32 OS and MsgBox Error is
thrown. After I click [OK] button, the message "Access is denied." is
printed on the console.
But I checked the executable image, it was compiled for a
32bit Intel386 CPU. Maybe compiler issues.
Re: Hannibal 1.7
Posted: Wed Aug 10, 2016 6:22 pm
by Edsel Apostol
What OS are you using? We don't have 32 bit OS anymore so that was not tested.
Re: Hannibal 1.7
Posted: Wed Aug 10, 2016 6:40 pm
by MikeGL
Edsel Apostol wrote:What OS are you using? We don't have 32 bit OS anymore so that was not tested.
Windows XP 32 bit.
pgado daan
I can try with 64 bit OS, but its not Windows.
