and though the board was complete, with no points marked differently, no way for a point to be marked differently, the score estimator said black wins by 0.5, but when going to scoring, it says white by 0.5 . Even after the game, clicking on score estimator will find black ahead by 0.5, when white has clearly won by 0.5.
I’m unable to find where the difference in scoring lies. I counted manually, and the result was, of course, correct, to have w ahead by .05.
Mainly i am just curious where the score estimator is seeing the extra point for black. I can’t figure it out!
I could be wrong about this but I seem to recall someone mentioning in a recent thread that the score estimator always uses Chinese scoring whereas your game was played with Japanese rules.
I’m a bit short on time right now so I haven’t checked your game myself but that is one possible source of the phantom point.
I too suspect this, but haven’t laboriously counted it out. As I understand it, Chinese and Japanese diverge if a player has played dead stones into the opponent’s territory and the opponent didn’t reply.
I had the same issue: https://online-go.com/game/13451200
The final score is 1.5 white wins, but estimator is 1.5 black wins. This is very misleading. The estimator should be using the same rules as the game or it make it worse than no estimator at all. Is there a way to file a bug report somewhere for this?
Thanks
The score estimator definitely needs some love from the devs. For reference, lots of little things are wrong with it:
it always uses Chinese rules;
it is not available in game reviews (badly needed);
you cannot click to mark groups dead or alive [1];
it has some display bugs [2] [3].
[1] Well, technically you can, but if the borders are not closed it just turns the entire board to one color or the other, which is useless (there is SHIFT-click, but it only marks stones dead, not alive).
[2] If you click the Estimate Score multiple times in a row (in an unfinished position), the estimated score changes each time.
[3] Open a finished game, click Estimate Score, then click Back to Game. The displayed score under the players’ names will be wrong.
I want to argue that there is in fact nothing wrong with the estimator: it is an estimator.
As an estimate, using Chinese or Japanese rules does not matter, they will be roughly the same in a usual game. It is meant to see in a very rough and inaccurate view which player is ahead. If the estimator says it is a close game, and you don’t see immediately that it is counting wrongly, I would suggest you start counting yourself, since having an accurate estimate of the score is then important for the game.
It is called an estimator for a reason that said, some improvements are indeed in order.
It does, and I am afraid adding other rulesets is not feasible at the moment. When it comes down to it, it would mean programming like 5 different estimators, which will occasionally have 1 or 2 points difference. That is usually meaningless to the estimate which can be 10 points off anyway until late endgame. There is also the fact that Chinese rules (I hear) are easiest to program and predict for some reason. Hence all bots count in Chinese.
Oh yes. It’s on the todo already.
Also very true, but not sure when that one is getting improved
Yes, but that’s actually intended and a good thing. (far be it for me to understand), but the estimator somehow uses partly random playouts. Meaning when you see the estimate is off, you can just click it again for maybe a better guess.
I am not an expert on rules, but it must the final one stone ko, no? For Chinese rules, filling the ko is an extra point (for the stone), while in Japanese it is not? But it is partly a guess.
How could it be otherwise? If it were otherwise, that would tell you that there is only one possible outcome of the game which would be boring wouldn’t it?
The different results give you the range of possible outcomes an estimator can guess at. That’s a good thing.
If you count the chinese score, you’ll find that black has 184 points. That translates into .5 win by black (assuming 6.5 komi), and the whole mystery boils down to black playing the last move, which is worth an extra point under chinese.
Imagine W had one more point of territory (instead of that last black move) - that would be +1 point for white under Japanese, but 2 point difference (+1 white, -1 black) under chinese, evening out the score.
AGA rules, which use Japanese-like scoring, counter this effect by ensuring white always plays (i.e. passes) last. In case of above game, W would need to pass twice (each pass being -1 point under AGA), evening the score with Chinese result again.
EDIT: Fun fact - in a game without a seki, under chinese rules and 6.5 komi, W+0.5 result is not possible. Do you know why?
It’s actually possible to mark a stone as dead by ctrl+clicking it. You can remove dead groups this way, though it often takes a few clicks to get the estimator to figure out that the group is actually dead. No way to mark groups as alive, though, AFAIK.
“AGA rules, which use Japanese-like scoring, counter this effect by ensuring white always plays (i.e. passes) last.”
Actually, the point of AGA rules is that you can either score a game the Chinese OR the Japanese way in the way you just described. However, on OGS, if AGA rules are chosen, the game is scored the Chinese way (no extra passes required at the end).
“If you count the chinese score […] (assuming 6.5 komi)”
Under Chinese rules, 7.5 komi is (almost always) used. This is the equivalent of 6.5 komi under Japanese rules. If you play a game with Chinese rules with 6.5 komi, that’s like playing a game with Japanese rules and 5.5 komi (if I understand correctly).
You are not trying to say that win/loss outcome of Chinese 7.5 komi game would be always equivalent to outcome of 6.5 komi Japanese scoring, are you? Because that wouldn’t be true. I’m not sure what picking a different komi changes in this context, because with any komi, the same subtle differences will appear.
Well, if you’re discontent with the SE, I suggest a single improvement.
Let users mark “imaginary boundaries” so the SE has an easier time estimating.
Whille you’re at it, after putting the imaginary boundaries in place, imagine black and white dots for black and white territory respectively, count intersections with a dead stone as 2 points, aggregate your scores, add the difference between captures, correct for komi, compute the difference and there you go, handmade score estimator.
“count intersections with a dead stone as 2 points, aggregate your scores, add the difference between captures”
Would only be necessary if the score estimator switched to using Japanese rules. Currently it uses Chinese rules, under which live stones are counted as territory - same result.
I know that most of the subtleties in my posts are so subtle at times as to elude even the distinguished reader, therefore I conjecture you may have missed the meaning of my post.
Obviously not, and it’s not that hard to find a counterexample. Imagine game with no prisoners, equal number of moves played by each player and B having 7 points advantage of territory on board
that said, some improvements are indeed in order.
It’s on the todo already.
