I focus on 9x9 games, due to short periods of available time. I’m around 10k. Mostly I play on another Go server where there are more players near my rank, which is why my profile shows few games here.
I’ve thought up a strategy for getting better at go where I start playing unranked against an AI bot, then at a critical move I somehow get the bot (or the OGS analysis AI) to give the move it would make if it were me.
If this is confusing, it just means that if I start as black, then at some point where it is black’s move, I can get an excellent playing suggestion from the bot or the analysis AI, as though the bot were playing as black.
After each such hint, the game play could continue to the end.
With this strategy, I can get to actual patterns of play, then discover from the bot’s feedback what the best move would be. Then I would look at the board and figure out why that move is considered to be best.
Currently, and often, I get to moves in the game where it is my turn, but I have only poor ideas of where to move. This strategy would help me learn what to do in such situations in real games.
So, two questions:
Is this a reasonable way to improve my game? If not, why not?
How can I do this in the OGS game interface? It isn’t obvious.
The short answer is - you CAN’T get this type of AI assistance from the OGS site while you are in the middle of an unfinished game. You can only get a review after a game is finished, showing you where your moves differed from what the AI would have played in that situation.
One of the main reasons the site doesn’t have this is that such a mechanism could be used to cheat by unscrupulous players.
Now obviously, you said you would only use this in unranked games, and only as a learning tool, but it’s important to keep in mind that there are other players out there that routinely get reported to the site admins for exactly this kind of AI-assist cheating, so please be aware that it is an issue out in the greater community.
If you would like this kind of assistance, I would recommend playing against an AI opponent on your computer or on other sites besides OGS. There are many options available:
These are AI tools you can install on your desktop computer, and they will have buttons you can enable / disable that will show you what the AI thinks are the best moves for that position. You can also upload SGF files of your own games, and get a move-by-move review showing AI suggestions. You can then play forward as many moves as you want from those positions (against the AI rather than your former opponent) to see how those alternate strategies may have worked out going forward.
ONLINE:
The one caveat I will mention here is that these are very high-level AI, so their play-style and level of expertise would be more appropriate to experienced SDK players.
As such, if you’re trying to find play-styles that are closer to your levels, it may be worth your time to find a lower-ranked AI bot like GnuGo (between 12 - 8 kyu), play games against that, and use KataGo to generate suggestions for your next moves
You can download a PandaNet SGF editor here which comes with a free version of GnuGo
So yeah, rather than messing around with AI-assisted play on OGS - which runs into all kinds of issues - it’s much easier to just do it on your computer.
If you want something quick, easy, and user-friendly, try this website
It runs in a browser, and it has a pretty broad range of functionality.
You can upload SGF files of your previous games, and then get move-by-move suggestions by hitting the [BEST] button.
You can play against the AI by hitting [AI PLAY] and then get suggestions for your moves
Also, rather than just giving you the “best move” suggestion, it gives you a range of suggestions (A, B, C, D, E etc) and shows you the potential score difference for each move, so you can compare different strategies from a given position.
If you’re still looking for something that lives on your desktop, you can always try LEELA
9x9 is often so sharp you may find the AI best move too hard to understand. Rather than trying to understand and play the AI best move, I would suggest instead to use it to identify and avoid your worst moves as you playing dumb moves is what makes you lose games, not the lack of brilliant moves. So after the game review yourself and think which moves were your worst or you had most difficulty with (thinking for yourself is key to improving), then ask an AI and find which ones lost more than 5 points say (AI sensei has tools like this), and look at the suggested better moves. Note you don’t have to play the blue -0 move instead, a green -0.3 points you understand is still loads better than a -5 points move and you are more likely to be able to do that better move in similar situations in future.
I know Leela supports 9x9 - it’s a bit less resource intensive than KaTrain
Still curious why KaTrain is not working on your machine. What OS are you using? Does your machine have enough memory to meet the requirements? Does it have a decent CPU / GPU?
Just to follow up that I’m now using KaTrain regularly, and finding it to supply exactly the kind of learning environment that I was sketching in my OP. The feature of showing influence as a detailed board shading is particularly helpful in visualizing weak areas of groups. I am very impressed by the variety of options and the overall flexibility of the analysis and display options. Everyone who wants to get better at Go should have the opportunity to learn about Katrain.exe for Windows.
If I could make one suggestion for KaTrain, it would be to have a way to speed up its response time, sacrificing accuracy. It would be nice to zip through a part of a game quickly, then restore full accuracy for the next portion of the game.
shinuito Just in case your negative post raises concerns about my ethics: this entire thread is about my learning Go using external tools OUTSIDE OF REGULAR GO GAMES WITH PEOPLE.
I completely disavow any negative implications imagined by shinuito .
I’m just quoting for clarity for others. I’m not trying to negatively imply anything against @david265. I don’t believe they’re doing anything wrong.
No Cheating or Computer Help
You can NEVER use Go programs (Leela, Zen, etc.) or neural networks to analyze current ongoing games unless specifically permitted (e.g., a computer tournament). The only type of computer assistance allowed is games databases for opening lines and joseki databases for corner patterns in correspondence Go. You cannot receive ANY outside assistance on live or blitz Go games.
In the ToS it doesn’t specify whether it’s ok in ranked or unranked, against bots or against humans.
I was clarifying that the interpretation I’m aware of is that if you’re playing unranked games, and the other player agrees, do what you like, play Gomoku, or knights move only Go etc.
And against bots, once it’s unranked I think it’s probably ok to do what you want against site bots. I think in the case the bot owner has issues with it, they can block the user from playing with that bot.