Thank you Werner. Good to hear you have managed to solve these problems. We'll see how it fares in the tests.Werner Taelemans wrote: ↑Wed Jun 09, 2021 8:33 am - The problem with time forfeits should be solved now.
- IID is back in.
- Galjoen doesn't use polling anymore. When it's not its turn, Galjoen should use zero CPU time.
- This should be the strongest version. At least in self play it is, although by a small margin.
Galjoen 0.41 released
Moderators: hgm, Rebel, chrisw
-
- Posts: 1362
- Joined: Sat Jul 21, 2018 7:43 am
- Location: Szentendre, Hungary
- Full name: Gabor Szots
Re: Galjoen 0.41.2 released
Gabor Szots
CCRL testing group
CCRL testing group
-
- Posts: 6991
- Joined: Thu Aug 18, 2011 12:04 pm
Re: Galjoen 0.41.2 released
You might want to try IIR instead of IID, it gave me 35 elo, it even gave stockfish 10 elo. It's just one instruction.Werner Taelemans wrote: ↑Wed Jun 09, 2021 8:33 am - The problem with time forfeits should be solved now.
- IID is back in.
- Galjoen doesn't use polling anymore. When it's not its turn, Galjoen should use zero CPU time.
- This should be the strongest version. At least in self play it is, although by a small margin.
http://www.talkchess.com/forum3/viewtop ... =7&t=74769
Meanwhile I have put Galjoen on the rack for the GRL, 700 games.
So far +37 elo.
http://rebel13.nl/a/grl.htm
Page refreshes each 30 seconds.
Results are kept, click Match History.
90% of coding is debugging, the other 10% is writing bugs.
-
- Posts: 119
- Joined: Mon Feb 03, 2014 11:57 am
- Location: Belgium
- Full name: Werner Taelemans
Re: Galjoen 0.41.2 released
That's interesting. It's also easy to implement. I started a test this afternoon, and after 200 games the result is still positive.Rebel wrote: ↑Wed Jun 09, 2021 10:54 am You might want to try IIR instead of IID, it gave me 35 elo, it even gave stockfish 10 elo. It's just one instruction.
http://www.talkchess.com/forum3/viewtop ... =7&t=74769
-
- Posts: 2555
- Joined: Fri Nov 26, 2010 2:00 pm
- Location: Czech Republic
- Full name: Martin Sedlak
Re: Galjoen 0.41.2 released
wow, Ed. how could I miss your IIR? running a test now and looks really good (after several failed patches in a row), I'll probably gain more than 10 elo at hyperbullet, which is actually quite a big gain for me.Rebel wrote: ↑Wed Jun 09, 2021 10:54 am You might want to try IIR instead of IID, it gave me 35 elo, it even gave stockfish 10 elo. It's just one instruction.
http://www.talkchess.com/forum3/viewtop ... =7&t=74769
the only drawback is that Cheng now fails to solve some tactical positions it used to, I'll probably try to find a cure and if I do without losing elo. It's always the same with elo, if it works in 99.9% positions, you gain elo. but sometimes I also care about the 0.1%
I hope you don't mind if I borrow your idea?
Martin Sedlak
-
- Posts: 6991
- Joined: Thu Aug 18, 2011 12:04 pm
Re: Galjoen 0.41.2 released
Sure, use it by all means if it produces. And it's true, a tactical loss in strength. But compensated in elo because of the increase in search depth. Also, at least in my case, it performed clearly better at longer time control. +20 at bullet, +35 at blitz.mar wrote: ↑Wed Jun 09, 2021 7:17 pmwow, Ed. how could I miss your IIR? running a test now and looks really good (after several failed patches in a row), I'll probably gain more than 10 elo at hyperbullet, which is actually quite a big gain for me.Rebel wrote: ↑Wed Jun 09, 2021 10:54 am You might want to try IIR instead of IID, it gave me 35 elo, it even gave stockfish 10 elo. It's just one instruction.
http://www.talkchess.com/forum3/viewtop ... =7&t=74769
the only drawback is that Cheng now fails to solve some tactical positions it used to, I'll probably try to find a cure and if I do without losing elo. It's always the same with elo, if it works in 99.9% positions, you gain elo. but sometimes I also care about the 0.1%
I hope you don't mind if I borrow your idea?
90% of coding is debugging, the other 10% is writing bugs.
-
- Posts: 1362
- Joined: Sat Jul 21, 2018 7:43 am
- Location: Szentendre, Hungary
- Full name: Gabor Szots
Re: Galjoen 0.41.2 released
Does that mean you are going to release a new version very soon?Werner Taelemans wrote: ↑Wed Jun 09, 2021 2:41 pmThat's interesting. It's also easy to implement. I started a test this afternoon, and after 200 games the result is still positive.Rebel wrote: ↑Wed Jun 09, 2021 10:54 am You might want to try IIR instead of IID, it gave me 35 elo, it even gave stockfish 10 elo. It's just one instruction.
http://www.talkchess.com/forum3/viewtop ... =7&t=74769
Gabor Szots
CCRL testing group
CCRL testing group
-
- Posts: 119
- Joined: Mon Feb 03, 2014 11:57 am
- Location: Belgium
- Full name: Werner Taelemans
Re: Galjoen 0.41.2 released
Not very soon.Gabor Szots wrote: ↑Fri Jun 11, 2021 8:08 am Does that mean you are going to release a new version very soon?
My latest tests at longer time controls are disappointing again.
Something else is going wrong and I need to find out what it is first.
-
- Posts: 1362
- Joined: Sat Jul 21, 2018 7:43 am
- Location: Szentendre, Hungary
- Full name: Gabor Szots
Re: Galjoen 0.41.2 released
Good luck solving the issue. In the meantime I have started a gauntlet with 0.41.2.Werner Taelemans wrote: ↑Fri Jun 11, 2021 10:02 amNot very soon.Gabor Szots wrote: ↑Fri Jun 11, 2021 8:08 am Does that mean you are going to release a new version very soon?
My latest tests at longer time controls are disappointing again.
Something else is going wrong and I need to find out what it is first.
Gabor Szots
CCRL testing group
CCRL testing group
-
- Posts: 119
- Joined: Mon Feb 03, 2014 11:57 am
- Location: Belgium
- Full name: Werner Taelemans
Re: Galjoen 0.41.2 released
Thanks.Gabor Szots wrote: ↑Fri Jun 11, 2021 3:57 pm Good luck solving the issue. In the meantime I have started a gauntlet with 0.41.2.
I found the cause of the problem. For years I've been using the same opening book to train and test Galjoen.
This book served me well in the past, but now, version 0.41 is totally overfitted on these openings.
When I use another book, I see an immediate drop in performance of at least 40 Elo.
I'll leave 0.41 for what it is, and I will use other openings to train the next version of Galjoen.
-
- Posts: 249
- Joined: Fri Mar 10, 2006 10:26 am
- Location: France
Re: Galjoen 0.41 released
Hi,
Is the version 0.41.2 downloadable somewhere, because I can only find the v. 0.41.1 on your site ?
Is the version 0.41.2 downloadable somewhere, because I can only find the v. 0.41.1 on your site ?