New engine releases 2020
Moderators: hgm, Rebel, chrisw
-
- Posts: 326
- Joined: Mon May 24, 2010 4:32 pm
Re: New engine releases 2020
Thanks very much Terje !! Congrats for this tremendous improvment !!
-
- Posts: 1362
- Joined: Sat Jul 21, 2018 7:43 am
- Location: Szentendre, Hungary
- Full name: Gabor Szots
Re: New engine releases 2020
After about 100 games against 2900 engines the score stands at 36 %, which shows that the improvement is closer to 150 than 250.
Gabor Szots
CCRL testing group
CCRL testing group
-
- Posts: 347
- Joined: Tue Nov 19, 2019 4:34 am
- Location: https://github.com/TerjeKir/weiss
- Full name: Terje Kirstihagen
Re: New engine releases 2020
That wouldn't be entirely unexpected, self-play elo is usually inflated. I had my hopes up after xr_a_y's test, as well as a (short) local test vs Glaurung 2.2 (2800 ccrl) which showed Weiss as 100 elo higher, though.Gabor Szots wrote: ↑Mon Apr 27, 2020 9:00 amAfter about 100 games against 2900 engines the score stands at 36 %, which shows that the improvement is closer to 150 than 250.
What time control btw?
-
- Posts: 1871
- Joined: Sat Nov 25, 2017 2:28 pm
- Location: France
Re: New engine releases 2020
Mine was 20+0.2 on an i7-9700KTerje wrote: ↑Mon Apr 27, 2020 11:44 amThat wouldn't be entirely unexpected, self-play elo is usually inflated. I had my hopes up after xr_a_y's test, as well as a (short) local test vs Glaurung 2.2 (2800 ccrl) which showed Weiss as 100 elo higher, though.Gabor Szots wrote: ↑Mon Apr 27, 2020 9:00 amAfter about 100 games against 2900 engines the score stands at 36 %, which shows that the improvement is closer to 150 than 250.
What time control btw?
-
- Posts: 1362
- Joined: Sat Jul 21, 2018 7:43 am
- Location: Szentendre, Hungary
- Full name: Gabor Szots
Re: New engine releases 2020
Gabor Szots
CCRL testing group
CCRL testing group
-
- Posts: 347
- Joined: Tue Nov 19, 2019 4:34 am
- Location: https://github.com/TerjeKir/weiss
- Full name: Terje Kirstihagen
Re: New engine releases 2020
Ah ok, the time management change had a much bigger impact on 10+0.1 than 60+0.6, and Weiss is probably stronger at (very) short TCs in general due to speed mattering more.
Out of curiousity, what types of adjudication did both of you use? I suspect Weiss might still mess up in end games that reach TB positions if there isn't adjudication. Though hopefully it doesn't run out of time anymore. (Please report if it does lose on time or shows any other unwanted behavior )
-
- Posts: 1871
- Joined: Sat Nov 25, 2017 2:28 pm
- Location: France
Re: New engine releases 2020
I always use no adjudication to watch end-games.Terje wrote: ↑Mon Apr 27, 2020 2:43 pmAh ok, the time management change had a much bigger impact on 10+0.1 than 60+0.6, and Weiss is probably stronger at (very) short TCs in general due to speed mattering more.
Out of curiousity, what types of adjudication did both of you use? I suspect Weiss might still mess up in end games that reach TB positions if there isn't adjudication. Though hopefully it doesn't run out of time anymore. (Please report if it does lose on time or shows any other unwanted behavior )
-
- Posts: 1362
- Joined: Sat Jul 21, 2018 7:43 am
- Location: Szentendre, Hungary
- Full name: Gabor Szots
Re: New engine releases 2020
I use three types of adjudication under Arena:Terje wrote: ↑Mon Apr 27, 2020 2:43 pm Out of curiousity, what types of adjudication did both of you use? I suspect Weiss might still mess up in end games that reach TB positions if there isn't adjudication. Though hopefully it doesn't run out of time anymore. (Please report if it does lose on time or shows any other unwanted behavior )
1. A 4-piece TB draw (so that e.g. KRKR endgames won't play out till the 50-moves rule takes effect. Not in force for weakish engines.
2. Games reaching 230 moves are adjudicated as a draw.
3. Resign threshold. For better engines I set this to 700 cp's, for weaker ones to 900.
For very weak engines I switch off all adjudications.
So for Weiss, 1-2-3 are all in effect.
I also use adjudications under Fritz and Shredder. I don't know the exact workings of those but I choose the most rigid ones (so that that game lasts longer), such as 'late' adjudication for Fritz.
Gabor Szots
CCRL testing group
CCRL testing group
-
- Posts: 584
- Joined: Fri Mar 30, 2018 7:20 am
- Full name: Andreas Matthies
Re: New engine releases 2020
Hi.
I received reports about Rubi 1.7.1 crashing sometimes and could watch it live crashing at the TCEC cup tournament yesterday so I released a fixed version here: https://github.com/Matthies/RubiChess/r ... /tag/1.7.2
Strength should be almost the same, so testers should use this new version but don't need to throw away games of 1.7.1.
Andreas
I received reports about Rubi 1.7.1 crashing sometimes and could watch it live crashing at the TCEC cup tournament yesterday so I released a fixed version here: https://github.com/Matthies/RubiChess/r ... /tag/1.7.2
Strength should be almost the same, so testers should use this new version but don't need to throw away games of 1.7.1.
Andreas
-
- Posts: 347
- Joined: Tue Nov 19, 2019 4:34 am
- Location: https://github.com/TerjeKir/weiss
- Full name: Terje Kirstihagen
Re: New engine releases 2020
I suppose Weiss is just better at very short time controls then. As always thanks for testing @Gabor, and also @xr_a_y