SF net ??? Should be banned from a tourney!
Build The Sky For My Dawn (Andscacs-Defen-Ethereal-Igel-Komodo-Leela-Minic-Nemori-Pedone-Rubi-Schooner-Slow-Toga-Wasp)
Moderators: hgm, Rebel, chrisw
-
- Posts: 288
- Joined: Sat Jun 30, 2018 10:58 pm
- Location: Ukraine
- Full name: Volodymyr Shcherbyna
Re: Build The Sky For My Dawn (Andscacs-Defen-Ethereal-Igel-Komodo-Leela-Minic-Nemori-Pedone-Rubi-Schooner-Slow-Toga-Was
-
- Posts: 585
- Joined: Fri Mar 30, 2018 7:20 am
- Full name: Andreas Matthies
Re: Build The Sky For My Dawn (Andscacs-Defen-Ethereal-Igel-Komodo-Leela-Minic-Nemori-Pedone-Rubi-Schooner-Slow-Toga-Was
The author clearly said that it is just an experiment and he didn't want to see AndscacsNNUE0.1 in any ranking. Same did I and some others in the past. The problem is that some people just ignore this and grab anything with NNUE label like the holy grail without further review.
Andreas
-
- Posts: 319
- Joined: Sat Oct 31, 2020 1:04 am
- Full name: Aleksey Glebov
Re: Build The Sky For My Dawn (Andscacs-Defen-Ethereal-Igel-Komodo-Leela-Minic-Nemori-Pedone-Rubi-Schooner-Slow-Toga-Was
Hi, guys!
I want to clarify some issues. First of all, it's my own private tourney and it's up to me to decide whom to put into it and whom to ban. Is this OK? Right now I'm curious about how strong new Andscacs with the (damned, provoking) SF NNUE net is. The tournament runs for (my personal) fun and its results doesn't go to any *official* rating list such as CCRL, CEGT, or FGRL. Furthermore. in general, I don't like the idea of paste-copying SF NNUE evals to other engines. In some 2-3 weeks I am going to organize another tourney which will include only NN(UE) engines and only with the original nets (at least not SF NNUE and not Leela nets). For this upcoming tourney I will not take Andscacs NNUE, BBC, Mayhem, and Tornado engines unless their authors train their own nets using their original data. By the similar reasons, I will include Nemorino 6.04_net_14 rather than Nemorino 6.00_net_6 (which is stronger but was trained using the SF data and architecture). Similarly, I will take Minic using one of the newest original nets (NiNy or NaNi) but not NaNe, NaNu, or NeNe even though the last 3 nets are much stronger. Finally, Toga III 0.2 net-0.5 will be included (which uses dkappe's toga 0.5-net and the original search algorithms from Toga II 4.0.1) rather than Toga III 0.1 NR which plays with toga net-0.4 but takes the advantages of unique and very powerful Stockfish's search algorithms that are implemented into the NetRunner.
I want to clarify some issues. First of all, it's my own private tourney and it's up to me to decide whom to put into it and whom to ban. Is this OK? Right now I'm curious about how strong new Andscacs with the (damned, provoking) SF NNUE net is. The tournament runs for (my personal) fun and its results doesn't go to any *official* rating list such as CCRL, CEGT, or FGRL. Furthermore. in general, I don't like the idea of paste-copying SF NNUE evals to other engines. In some 2-3 weeks I am going to organize another tourney which will include only NN(UE) engines and only with the original nets (at least not SF NNUE and not Leela nets). For this upcoming tourney I will not take Andscacs NNUE, BBC, Mayhem, and Tornado engines unless their authors train their own nets using their original data. By the similar reasons, I will include Nemorino 6.04_net_14 rather than Nemorino 6.00_net_6 (which is stronger but was trained using the SF data and architecture). Similarly, I will take Minic using one of the newest original nets (NiNy or NaNi) but not NaNe, NaNu, or NeNe even though the last 3 nets are much stronger. Finally, Toga III 0.2 net-0.5 will be included (which uses dkappe's toga 0.5-net and the original search algorithms from Toga II 4.0.1) rather than Toga III 0.1 NR which plays with toga net-0.4 but takes the advantages of unique and very powerful Stockfish's search algorithms that are implemented into the NetRunner.
Incredibly fast systems miscount incredibly fast.
-
- Posts: 585
- Joined: Fri Mar 30, 2018 7:20 am
- Full name: Andreas Matthies
Re: Build The Sky For My Dawn (Andscacs-Defen-Ethereal-Igel-Komodo-Leela-Minic-Nemori-Pedone-Rubi-Schooner-Slow-Toga-Was
Completely okay. You may noticed that my first answer went to Frank, not to you.
BTW. Did you see my question in your YT broadcast?
Andreas
-
- Posts: 6808
- Joined: Wed Nov 18, 2009 7:16 pm
- Location: Gutweiler, Germany
- Full name: Frank Quisinsky
Re: Build The Sky For My Dawn (Andscacs-Defen-Ethereal-Igel-Komodo-Leela-Minic-Nemori-Pedone-Rubi-Schooner-Slow-Toga-Was
Completely OK from my side.
Keep up your good work ... feel free to add in your tournaments whatever you like.
I will follow, you can be sure ... with many interest, I like your work!!
Best
Frank
Keep up your good work ... feel free to add in your tournaments whatever you like.
I will follow, you can be sure ... with many interest, I like your work!!
Best
Frank
-
- Posts: 319
- Joined: Sat Oct 31, 2020 1:04 am
- Full name: Aleksey Glebov
Re: Build The Sky For My Dawn (Andscacs-Defen-Ethereal-Igel-Komodo-Leela-Minic-Nemori-Pedone-Rubi-Schooner-Slow-Toga-Was
I don't see any comments/questions on my Youtube Broadcast. Where did you leave it? May be you can ask your question here?
Incredibly fast systems miscount incredibly fast.
-
- Posts: 585
- Joined: Fri Mar 30, 2018 7:20 am
- Full name: Andreas Matthies
Re: Build The Sky For My Dawn (Andscacs-Defen-Ethereal-Igel-Komodo-Leela-Minic-Nemori-Pedone-Rubi-Schooner-Slow-Toga-Was
Question was about which binary/binaries you use for the two different Rubi versions playing.
I would expect the 0112 net playing better than the 1107 net (just did another test here and it scored ~ +15Elo). So the current opposite result may be
1. small sample size
2. binary not optimal
3. regression for some unknown reason
Regards, Andreas
-
- Posts: 319
- Joined: Sat Oct 31, 2020 1:04 am
- Full name: Aleksey Glebov
Re: Build The Sky For My Dawn (Andscacs-Defen-Ethereal-Igel-Komodo-Leela-Minic-Nemori-Pedone-Rubi-Schooner-Slow-Toga-Was
Both nets run with RubiChess-1.9-BMI2 compile which was the fastest for my system with NN-1107 net. I didn't make a new speed test for NN-0112 net and used the same compile. May be that was a mistake?
Incredibly fast systems miscount incredibly fast.
-
- Posts: 585
- Joined: Fri Mar 30, 2018 7:20 am
- Full name: Andreas Matthies
Re: Build The Sky For My Dawn (Andscacs-Defen-Ethereal-Igel-Komodo-Leela-Minic-Nemori-Pedone-Rubi-Schooner-Slow-Toga-Was
Should be okay. This is what I tested with +15 Elo for 0112. So it is probably just small sample size.
-
- Posts: 319
- Joined: Sat Oct 31, 2020 1:04 am
- Full name: Aleksey Glebov
Re: Build The Sky For My Dawn (Andscacs-Defen-Ethereal-Igel-Komodo-Leela-Minic-Nemori-Pedone-Rubi-Schooner-Slow-Toga-Was
Incredibly fast systems miscount incredibly fast.