Feature idea: Report on all "Losing Moves" you've made from the AI's POV

As I’ve been getting into 9x9 over the last few months I’ve started to suspect there’s some patterns to games I’m losing and thought it’d be interesting to have a report on moves (across all your games) that lead to a significant loss in win percentage from the view of the AI. If it’s possible it’d provide some nice insight into repeated mistakes over time. Anyone else think this might be a good idea? Is it already out there and I’m just missing it?

Thanks!

(per: Vsotvep below, edited to clarify this is a suggested report across all games since we already have in-game reporting)

I’m unsure if I understand what you mean, but you can read out your largest mistakes from the AI review graph (if you don’t see the graph, make sure you have AI review enabled in the right hand menu next to the game):

This is the graph of your last game. As you can see, your opponent gained a small advantage after move 4, there were three consecutive moves, 16, 18 and 20, where you missed the biggest spot to play, and the move that lost you the game was move 24. Meanwhile, if you look at the score graph, you see that you lost a couple of extra points at move 30 and move 36 as well.

Note that you can click at any point in the graph to be moved towards that move and inspect what the AI suggests as an alternative.

2 Likes

Yes Vsotvep I use review graphs every game and they’re a big factor in my improving, awesome feature! What I’m imagining here is a report across all of your games compiling (for example) everything as bad as my move 24 above, so that you can see if you make a certain type of misplay a lot and adjust accordingly.

I can see my original post was unclear on the “across all games” point so thanks for helping clarify Vsotvep.

Ah, I get it now.

Off the top of my head, it will be quite a hard thing to make, since the AI does not give any kind of assessment on what the type of mistake is that you’re making. Unless it’s a joseki mistake, I’m not sure if the position in many games can be compared that easily by a computer, and we’d probably need specialised software for something like that.

1 Like

The easiest thing I can imagine is for the AI to log in a separate table every move you make with a >90% loss impact, and then to provide a report of those. From there you could review these losing moves over time and hopefully draw the necessary connections. It does sound like I’m suggesting some extra work but I’d find this really helpful, and not to mention interesting!

Can’t you just bookmark each game after you played them where there is a big drop? Since you will be the most qualify to know the context of those games (and what constitute a “big mistake”). And just click those bookmark game links to review each of them.

claire_yang I do that to some extent, but I think automating cross-game data would make the AI even more useful and speed up learning.

Still not clear what cross-game data means. Different games have different context, how do you wish them to be “compiled together”? I need some kind of examples.

I think that only a human stronger player could look at your games and find common problems in your style of play.

You don’t need an AI for answering your question: you need a teacher. :slightly_smiling_face:

5 Likes

Just my humble opinion, but I think this is a niche enough use case that I don’t think it would become a feature. A couple of ideas though:

  1. If you are technically proficient, I would recommend using the REST API to download your games, and possibly run them through a stronger AI on your machine. User games available at https://online-go.com/api/v1/players/{YOUR_PLAYER_ID}/games

  2. If you are less technically proficient, you could just record important moves in an excel sheet and go over them later. This seems like a lot of work, but determining “best” moves is not easily automated anyway, so it would be good to hand-pick the moves you are looking for.

1 Like

+1 A pro reviewing 3 games >>> me reviewing a massive corpus by myself :slight_smile:

2 Likes

claire_yang: The AI rates all of your moves in terms of their impact on score and win percentage. I’m talking about compiling all of your moves, across all games, together based on that rating, e.g. a report that shows you every move you’ve made that caused you to drop 90% in terms of your chance of winning the game. From there you can go and analyze and look for common patterns.

lys: I’ve actually moved up 3-4kyu over the last few months and I think the OGS AI has helped a lot (as opposed to GoQuest where I would just get walloped and not get much insight). Getting a teacher’s perspective is of course a great idea, but the AI data is all there already, I’m suggesting a new way of making it accessible.

benjita: Using the REST API looks like a good way to handle this and I’ll probably take a crack at it, but I also think it’d be a cool feature with a (relatively) low lift that would benefit non-technical players too.

1 Like

You can download SGF from your game’s page and open them with LizGoban.
The software is very simple to install (just download, unzip and it’s done) and has lovely emoticons as stones, with shades of black and white to indicate influence of each group.

No need to be technically proficient to run katago on your pc. I use it sometimes on my laptop and worķs great!

OP wants to analyze games en-masse. If you want to download your entire game list quickly, the rest API is probably the best way to do it, but you need to be technically proficient.

For a one off game, yes, downloading the SGF and analyzing is easy to do!

1 Like