Very new to the game less than one week in I enjoy the game but also feeling like im pounding my head into a wall I don’t understand the rank systems in any real depth, but I do understand the 8k bot and the 1d destroy me every single game would it be possible to have a bot that doesn’t win by 50 - 100 points or more every game? would add a lot more enjoyment to the learning process.
Hello, it is a lengthy topic, but I will try to make my answer readable
similar idea has been talked about here - maybe some of it will also apply to you
First off - I would pretty much recommend getting over the anxiety and playing humans… We all started one day and most of us still remember the pain and are welcoming to beginners. Feel free to ask for a teaching game here at the forums.
OGS is not running any of the bots avilable and we are developing none. They are all managed by third parties and our users. Master mantis used to be avilable for beginners (although it was still about 15k)
now I see it is not listed. @Kungfu_Panda (sorry to bother, just tagging you to show we miss your bot )
But other than that it seems there is little interest in developing “bad” bots. If you insist on playing “amateur” level bots, there are some pretty basic on android devices:
oh also. As long as it is Kyu the lower the number the higher the skill. - amateurs start at 25k/30k and climb their way to 1k which is almost a decent level.
Once you are better then Kyus you become a Dan (a master level) then it goes the other way. The higher the Dan the better the skill.
(professional players have sort of their own ranks - but you can still think about tham as Dans - it is the same idea., just even stronger…)
In have absolutely nothing useful to say.
I just want to mention that i misread topic as “Boobs for noobs?” and got interested enough to click this open
Well since i’m already here, here’s my two cents about playing/training/learning against bots in general. Still not answering your question though…
You clickbaited yourself. Watch out, next thing you know, you’re selling yourself a car.
I started against “Go free” on Android. It has many levels of strenght and allows to rollback and try another sequence.
It helped me a lot when I was learning very basic notions.
I think it’s more useful than playing against a bot that you can’t control.
The free version has 9x9 and 13x13 boards. You can play on a 19x19 board only a limited number of moves.
I started against GnuGo, on a 9x9. Increase the handicap until you can win, which will get reasonably easy to do as you approach 9 stones. (If you’re having trouble, just focus on trying to connect all of your stones. This will result in having large, thick groups that will give you profit.) Mostly, you want to get used to recognizing things like “my stones are in atari”, or “my opponent can atari me in one move.” Until you can do those somewhat reliably, you’ll probably get more out of 9x9 than 19x19, where not only do all of those simple mechanical bits matter, but you need to pay some attention to full-board strategy. I’d recommend playing on the 9x9 board, at least until you can win on a 4 stone handicap, or so. At that point, try some 9 handicap games on 19x19 with the same strategy: try to connect your stones, and make good, strong shapes.
I think you can also download GnuGo (maybe it’s packaged with the PandaNet client?), and it has/had an adjustable difficulty level, but I’d recommend against playing a 25k bot. They tend to make lots of moves that are not only bad, but really weird. They won’t play bad “human” moves, and IMO aren’t a good learning resource. Playing a stronger bot, like the OGS version of GnuGo, and trying to get a decent result at a high handicap, will teach you more about shape and connection.
You can play a handicapped game. This is what face-to-face players of different ranks do.
Generally speaking, play a fair game against the AI. After the result, for every 15-30 points you lose by (others might have a more precise stone count) increase your handicap stones by 1 on the next game.
All the handicap does is let the player place a set number of stones on the board on their first turn when the game starts, so they get a bit of an advantage in different areas of the board.
That said, the speed in which a AI plays can be “intimidating” compared to the pace of regular players, but increasing the move time just lets it run deeper analysis.
The best bot for noobs is probably LeelaZero. Here’s how to get it running:
Download and extract LeelaZero.
Download the network file that matches your level to the directory where LeelaZero is.
- 30k: network 9
- 29k: network 10
- 28k: network 11
- 27k: network 12
- 26k: network 13
- 24k: network 14
- 23k: network 15
- 21k: network 16
- 20k: network 17
Download a Go client that supports a Go-playing program.
Connect LeelZero to the Go client.
Add parameters like the following: -g --noponder -p 100 -w [network filename goes here].gz
If anyone can point me to a game:
(weak) Leela Vs beginner
I am curious to see how she plays at low level.
See old match games here: http://zero.sjeng.org/network-profiles
For example, 400 games of network 12 vs. network 13.
I didn’t check the actual games, but just scanned the results. It quite interesting that resigns did not occur until game 221 out of 400, but then after that they become quite frequent. Is this series of 400 part of the training (or just the evaluation)? Is this the point that the bot learned to evaluate the game confidently enough to resign?
That’s really interesting thanks. Gave me already some ideas what she would do.
Now if any ddk did record here one of his game (well let’s dream) I would be pleased to see her against a human too.