I see what appears to be a new chart at the bottom of a finished game. This lists categories of moves, how often they came up, and (if you click) the move numbers themselves. Cool!
Personally, I welcome more tools for post-game analysis baked right into OGS.
How does everyone else feel about this feature?
I think its bad terminology to call “Joseki” something that is worse than “Good”, it will confuse everyone
People will suspect someone to use AI and then would use this tool to become sure.
Everyone would have their own different pseudoscience interpretation of this thing.
It would be witch hunt.
It’s a good feature if the goal is making money, because it makes me want to become an OGS supporter to see which moves were inaccurate. without having to plug my game into another software like ai-sensei.com or Katrain.
It’s easier to make inaccurate moves if the opponent is stronger, isn’t that what it means? It doesn’t mean the info is useless. It only means the info is not easy to use. If I play against the same opponent or bot (like my favorite, drunken master bot) 20 times in a month, this info can be a semi-consistent measure of performance. Of course you can’t use it to compare performance between playing vs. a 30 kyu and playing vs. Shin Jinseo.
Fun feature. Seeing these kinds of stats is one of the reasons I like putting games into Katrain. Pretty cool.
Indeed bizarre. A move being joseki is totally separate to how many points/percentage points it loses compared AI optimal move. There is a strong correlation, in that joseki moves will usually be good to very good, but they can be bad depending on the situation, though being blunder level should be very rare.
I wonder how the move is deemed to be “joseki”. I mean, joseki is always changing. Today’s joseki may not be joseki tomorrow.
Joseki moves are always excellent/great according to the AI but they are also not as interesting. I stuck them between good and inaccurate because they are moves players study and aren’t so much “good” or “bad” because one doesn’t really need to think about them so much, they’re just joseki, moves that are memorized as being good ways to start. Another alternative would be to move them out of the table entirely as they don’t really belong in the same sense as the other moves there, but I felt that would also be confusing. If anyone has any better ideas for presentation I’m all ears, but I don’t think they should be above or included in the “excellent” category, even though they are excellent moves according to the AI.
The system works by looking at sequences of good moves in the corner regions, if you explore joseki you’ll see many sequences that have similar score losses and predicted win rates.
What is joseki and what isn’t is debatable for a lot of sequences and situations. While I’m positive there’s going to be ways of improving the current algorithm that flags a move as joseki or not, I’d say the current flagging mostly agrees with known widely accepted patterns, and certainly includes some cases that wouldn’t be traditionally be called joseki, but might be worth considering or at least noted as interesting that the AI and algorithm said “hey that’s a pretty decent opening sequence”.
I’m definitely influenced by chess tools here, but it seems to me “how good were my moves?” and “how did my play relate to known joseki?” are separate questions. Or to rephrase slightly, move quality and move interestingness are two different axes of measurement, and including ‘joseki’ in the list (instead of just counting those moves as whatever quality level they are, and pulling more detailed joseki stats separately) muddies the two.
I do think it’d be interesting to dive more deeply into the joseki being recognized. How “on-book” are you, in terms of playing more common vs less common joseki?
love, love, love
Agrre, but the issue is in chess is more easy to do this with AIs. Your move just compares against the ai move over a “bar” of losing value, and a “book”/opening move, is just following the start of the game. But in Go there are so so much possibilities (including tenuki or playing in other side in sente and coming back) that is hard to do the same, should be something similar adapted. On the other side, a 1 point loss (a pawn) in chess it’s important because game trends to simplify, but Go 1 point (a pawn) compared to AI move is misery unles you are Pro level. AI in chess is usefull in almost all levels, in Go AI if you don’t know how to use it properly could be counterproductive
Would you rather that move be marked as “inaccuracy” or “good” or “excellent”? I think it’s more useful to place in “joseki” after reading anoek’s explanation. After all, it’s not that bad. Almost any first two or even three opening moves on 19x19 are good unless they’re on the 1st line. Only if you make a full square in the middle of nowhere then it starts being bad.
As a player, I find it surprising to be told that joseki moves are “not interesting”.
From a player perspective, they certainly seem interesting to me - I struggle with them just as much as any other move - even if perhaps we can’t take as much credit for them since we didn’t have to find them out of all possible moves. So you could say we did not do “such a good job”.
So is the move not excellent then, just because “we didn’t think of it on our own”?
We still had to choose which joseki based on the whole board, and remember it.
I think I agree with the poster who said “interesting” is a different axis to “quality”.
An excellent move is excellent regardless of who thought of it.
As an OJE curator, I also wonder whether this will force us to accept any move made by the AI in a corner sequence as “Joseki”? “Is AI play an acceptable source for Joseki?” has been an ongoing bone of contention…
It’s true, and if the general sentiment is that we shouldn’t carve those out into their own category that’s easy enough to do. I guess as I was working through development I found myself skipping past the first few “excellent” moves because for instance Q16 R14 O17 are excellent, but to me not interesting. Repeat that and similar very common patterns for all of the corners and I felt there was a lot of noise in the “Excellent” category that are technically good moves, but there wasn’t anything to learn from them.
I wonder if “Joseki” moves should just be pulled out, and counted in a separate “set”, not placed relative to the others in the scale - it would show is that "this number of moves were not evaluated on the “Excellent-Blunder scale because you didn’t think of them, you just got them from memory”.
This still doesn’t feel right becase a Joseki move can be a blunder can’t it? If you choose one where you don’t have the ladder … it’s both “Joseki” and “Blunder”.
Actually, its an interesting question about the definition of Joseki. If you’re watching play and you see someone play a move out of a Joseki dictionary, but you know that the commentary of the Joseki says “if black has the ladder”, then do you say “Joseki” as they place the stone even though they don’t have the ladder? Or is it “not a joseki move” precisely because they don’t have the ladder?
@anoek The only critique I have of the feature is the UI design of the popover element:
I’m not a graphic designer but I’ve done some frontend dev before and fiddling with the CSS I feel like this looks a little bit cleaner (but still nowhere near perfect) and easier to press on mobile devices:
There should be a way to align the popover with the side of the container/anchor div in the UI library. It looks weird to me floating in the middle on large screens, I don’t know.
