Closed justremember closed 3 years ago
Please consider this as a priority
There was a bug in the configuration of the current's bot difficulty, causing it to play below what it could. That should be fixed since yesterday. Did you notice improvements in the difficulty?
Haven't tested it myself but Illya, a 1 dan in 81dojo, just beat AI level 8 an hour ago. https://lishogi.org/p38HB5gU
You can also play against it on https://www.pychess.org/ to compare. The difficulty settings there should be identical, so if the actual playing strength differs, there might be something off with the settings. Normally, level 8 should be very tough for a 1 dan (speaking as a 1 kyu FESA myself). Someone rated ~2000 on 81dojo also played dozens of games against it on pychess and he only won a few of them.
I guess set up a engine is not just to play against it, also to analize games, there it should be a very good one. By now If I want to analize with computer a chess game I have just one click in lichess, but in 81dojo I have to export kifu, set up shogiGui and import there, as my PC is old, I can't support yanegura OU, other options is use kento, but everything is more and more steps. Sadly stockfish is not a profesional shogi player to help me to improve, I tried several times and just can say it is like play against a shogi game for a old Nintendo.
El lun., 23 de nov. de 2020 4:43 AM, Fabian Fichter < notifications@github.com> escribió:
You can also play against it on https://www.pychess.org/ to compare. The difficulty settings there should be identical, so if the actual playing strength differs, there might be something off with the settings. Normally, level 8 should be very tough for a 1 dan (speaking as a 1 kyu FESA myself). Someone rated ~2000 on 81dojo also played dozens of games against it on pychess and he only won a few of them.
— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/WandererXII/lishogi/issues/66#issuecomment-731985753, or unsubscribe https://github.com/notifications/unsubscribe-auth/AGMYMBM2IN3L6WMV35QEDJDSRIHBVANCNFSM4T4G42WA .
Sure, switching to a stronger engine of course makes sense (especially for saving server resources), and I am by no means arguing against that, but just providing input to identify potential issues with the settings which can still be useful until the switch will eventually be done.
From my point of view the following points might be relevant for this topic:
Skill Level
parameter and use the USI protocol, so using a different engine for (not yet available) server-side analysis as well as game play should be relatively easy.https://lishogi.org/U2K2pvvp/black if I can win against SF level 8, it is not strong, Im a kyu level player (seeking for shodan)
Does anyone know what the engine is behind the computer player of PlayShogi.com? It is very fast and very strong, stronger than Bonanza on my local machine.
The current LiChess computer player is doing the community a disservice, as it is not demonstrating good shape or joseki at all.
Pretty sure it's yaneuraou + elmo. Having fairy stockfish for local evaluation is better than none at all, it can detect mates and tactics, but I believe users should be made aware its materialistic, lacks an eye on strategy, and is (a lot?) weaker than the strongest amateurs.
If yaneuraou for AI opponent does get implemented, I think having more levels than just 8 is appropriate. Maybe 20.
Right now, fairy stockfish is much weaker than current NNs. It would be nice to play against a decent bot (and in the future, have it analyze my games from the site itself). But the main concern I think is how NNs are currently resource intensive. An NN with less resources could still be stronger than fairy though.
This is a docker setup that installs yaneuraou + orqha/elmo alongside shogigui, could help: https://github.com/jruffet/docker-shogigui